xref: /openbmc/qemu/hw/ppc/spapr.c (revision 83a71719)
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 "qemu/cutils.h"
74 #include "hw/ppc/spapr_cpu_core.h"
75 #include "hw/mem/memory-device.h"
76 
77 #include <libfdt.h>
78 
79 /* SLOF memory layout:
80  *
81  * SLOF raw image loaded at 0, copies its romfs right below the flat
82  * device-tree, then position SLOF itself 31M below that
83  *
84  * So we set FW_OVERHEAD to 40MB which should account for all of that
85  * and more
86  *
87  * We load our kernel at 4M, leaving space for SLOF initial image
88  */
89 #define FDT_MAX_SIZE            0x100000
90 #define RTAS_MAX_SIZE           0x10000
91 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
92 #define FW_MAX_SIZE             0x400000
93 #define FW_FILE_NAME            "slof.bin"
94 #define FW_OVERHEAD             0x2800000
95 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
96 
97 #define MIN_RMA_SLOF            128UL
98 
99 #define PHANDLE_XICP            0x00001111
100 
101 /* These two functions implement the VCPU id numbering: one to compute them
102  * all and one to identify thread 0 of a VCORE. Any change to the first one
103  * is likely to have an impact on the second one, so let's keep them close.
104  */
105 static int spapr_vcpu_id(sPAPRMachineState *spapr, int cpu_index)
106 {
107     assert(spapr->vsmt);
108     return
109         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
110 }
111 static bool spapr_is_thread0_in_vcore(sPAPRMachineState *spapr,
112                                       PowerPCCPU *cpu)
113 {
114     assert(spapr->vsmt);
115     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
116 }
117 
118 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
119 {
120     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
121      * and newer QEMUs don't even have them. In both cases, we don't want
122      * to send anything on the wire.
123      */
124     return false;
125 }
126 
127 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
128     .name = "icp/server",
129     .version_id = 1,
130     .minimum_version_id = 1,
131     .needed = pre_2_10_vmstate_dummy_icp_needed,
132     .fields = (VMStateField[]) {
133         VMSTATE_UNUSED(4), /* uint32_t xirr */
134         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
135         VMSTATE_UNUSED(1), /* uint8_t mfrr */
136         VMSTATE_END_OF_LIST()
137     },
138 };
139 
140 static void pre_2_10_vmstate_register_dummy_icp(int i)
141 {
142     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
143                      (void *)(uintptr_t) i);
144 }
145 
146 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
147 {
148     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
149                        (void *)(uintptr_t) i);
150 }
151 
152 int spapr_max_server_number(sPAPRMachineState *spapr)
153 {
154     assert(spapr->vsmt);
155     return DIV_ROUND_UP(max_cpus * spapr->vsmt, smp_threads);
156 }
157 
158 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
159                                   int smt_threads)
160 {
161     int i, ret = 0;
162     uint32_t servers_prop[smt_threads];
163     uint32_t gservers_prop[smt_threads * 2];
164     int index = spapr_get_vcpu_id(cpu);
165 
166     if (cpu->compat_pvr) {
167         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
168         if (ret < 0) {
169             return ret;
170         }
171     }
172 
173     /* Build interrupt servers and gservers properties */
174     for (i = 0; i < smt_threads; i++) {
175         servers_prop[i] = cpu_to_be32(index + i);
176         /* Hack, direct the group queues back to cpu 0 */
177         gservers_prop[i*2] = cpu_to_be32(index + i);
178         gservers_prop[i*2 + 1] = 0;
179     }
180     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
181                       servers_prop, sizeof(servers_prop));
182     if (ret < 0) {
183         return ret;
184     }
185     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
186                       gservers_prop, sizeof(gservers_prop));
187 
188     return ret;
189 }
190 
191 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
192 {
193     int index = spapr_get_vcpu_id(cpu);
194     uint32_t associativity[] = {cpu_to_be32(0x5),
195                                 cpu_to_be32(0x0),
196                                 cpu_to_be32(0x0),
197                                 cpu_to_be32(0x0),
198                                 cpu_to_be32(cpu->node_id),
199                                 cpu_to_be32(index)};
200 
201     /* Advertise NUMA via ibm,associativity */
202     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
203                           sizeof(associativity));
204 }
205 
206 /* Populate the "ibm,pa-features" property */
207 static void spapr_populate_pa_features(sPAPRMachineState *spapr,
208                                        PowerPCCPU *cpu,
209                                        void *fdt, int offset,
210                                        bool legacy_guest)
211 {
212     uint8_t pa_features_206[] = { 6, 0,
213         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
214     uint8_t pa_features_207[] = { 24, 0,
215         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
216         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
217         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
218         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
219     uint8_t pa_features_300[] = { 66, 0,
220         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
221         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
222         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
223         /* 6: DS207 */
224         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
225         /* 16: Vector */
226         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
227         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
228         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
229         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
230         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
231         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
232         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
233         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
235         /* 42: PM, 44: PC RA, 46: SC vec'd */
236         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
237         /* 48: SIMD, 50: QP BFP, 52: String */
238         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
239         /* 54: DecFP, 56: DecI, 58: SHA */
240         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
241         /* 60: NM atomic, 62: RNG */
242         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
243     };
244     uint8_t *pa_features = NULL;
245     size_t pa_size;
246 
247     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
248         pa_features = pa_features_206;
249         pa_size = sizeof(pa_features_206);
250     }
251     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
252         pa_features = pa_features_207;
253         pa_size = sizeof(pa_features_207);
254     }
255     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
256         pa_features = pa_features_300;
257         pa_size = sizeof(pa_features_300);
258     }
259     if (!pa_features) {
260         return;
261     }
262 
263     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
264         /*
265          * Note: we keep CI large pages off by default because a 64K capable
266          * guest provisioned with large pages might otherwise try to map a qemu
267          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
268          * even if that qemu runs on a 4k host.
269          * We dd this bit back here if we are confident this is not an issue
270          */
271         pa_features[3] |= 0x20;
272     }
273     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
274         pa_features[24] |= 0x80;    /* Transactional memory support */
275     }
276     if (legacy_guest && pa_size > 40) {
277         /* Workaround for broken kernels that attempt (guest) radix
278          * mode when they can't handle it, if they see the radix bit set
279          * in pa-features. So hide it from them. */
280         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
281     }
282 
283     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
284 }
285 
286 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
287 {
288     int ret = 0, offset, cpus_offset;
289     CPUState *cs;
290     char cpu_model[32];
291     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
292 
293     CPU_FOREACH(cs) {
294         PowerPCCPU *cpu = POWERPC_CPU(cs);
295         DeviceClass *dc = DEVICE_GET_CLASS(cs);
296         int index = spapr_get_vcpu_id(cpu);
297         int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
298 
299         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
300             continue;
301         }
302 
303         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
304 
305         cpus_offset = fdt_path_offset(fdt, "/cpus");
306         if (cpus_offset < 0) {
307             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
308             if (cpus_offset < 0) {
309                 return cpus_offset;
310             }
311         }
312         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
313         if (offset < 0) {
314             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
315             if (offset < 0) {
316                 return offset;
317             }
318         }
319 
320         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
321                           pft_size_prop, sizeof(pft_size_prop));
322         if (ret < 0) {
323             return ret;
324         }
325 
326         if (nb_numa_nodes > 1) {
327             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
328             if (ret < 0) {
329                 return ret;
330             }
331         }
332 
333         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
334         if (ret < 0) {
335             return ret;
336         }
337 
338         spapr_populate_pa_features(spapr, cpu, fdt, offset,
339                                    spapr->cas_legacy_guest_workaround);
340     }
341     return ret;
342 }
343 
344 static hwaddr spapr_node0_size(MachineState *machine)
345 {
346     if (nb_numa_nodes) {
347         int i;
348         for (i = 0; i < nb_numa_nodes; ++i) {
349             if (numa_info[i].node_mem) {
350                 return MIN(pow2floor(numa_info[i].node_mem),
351                            machine->ram_size);
352             }
353         }
354     }
355     return machine->ram_size;
356 }
357 
358 static void add_str(GString *s, const gchar *s1)
359 {
360     g_string_append_len(s, s1, strlen(s1) + 1);
361 }
362 
363 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
364                                        hwaddr size)
365 {
366     uint32_t associativity[] = {
367         cpu_to_be32(0x4), /* length */
368         cpu_to_be32(0x0), cpu_to_be32(0x0),
369         cpu_to_be32(0x0), cpu_to_be32(nodeid)
370     };
371     char mem_name[32];
372     uint64_t mem_reg_property[2];
373     int off;
374 
375     mem_reg_property[0] = cpu_to_be64(start);
376     mem_reg_property[1] = cpu_to_be64(size);
377 
378     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
379     off = fdt_add_subnode(fdt, 0, mem_name);
380     _FDT(off);
381     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
382     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
383                       sizeof(mem_reg_property))));
384     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
385                       sizeof(associativity))));
386     return off;
387 }
388 
389 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
390 {
391     MachineState *machine = MACHINE(spapr);
392     hwaddr mem_start, node_size;
393     int i, nb_nodes = nb_numa_nodes;
394     NodeInfo *nodes = numa_info;
395     NodeInfo ramnode;
396 
397     /* No NUMA nodes, assume there is just one node with whole RAM */
398     if (!nb_numa_nodes) {
399         nb_nodes = 1;
400         ramnode.node_mem = machine->ram_size;
401         nodes = &ramnode;
402     }
403 
404     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
405         if (!nodes[i].node_mem) {
406             continue;
407         }
408         if (mem_start >= machine->ram_size) {
409             node_size = 0;
410         } else {
411             node_size = nodes[i].node_mem;
412             if (node_size > machine->ram_size - mem_start) {
413                 node_size = machine->ram_size - mem_start;
414             }
415         }
416         if (!mem_start) {
417             /* spapr_machine_init() checks for rma_size <= node0_size
418              * already */
419             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
420             mem_start += spapr->rma_size;
421             node_size -= spapr->rma_size;
422         }
423         for ( ; node_size; ) {
424             hwaddr sizetmp = pow2floor(node_size);
425 
426             /* mem_start != 0 here */
427             if (ctzl(mem_start) < ctzl(sizetmp)) {
428                 sizetmp = 1ULL << ctzl(mem_start);
429             }
430 
431             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
432             node_size -= sizetmp;
433             mem_start += sizetmp;
434         }
435     }
436 
437     return 0;
438 }
439 
440 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
441                                   sPAPRMachineState *spapr)
442 {
443     PowerPCCPU *cpu = POWERPC_CPU(cs);
444     CPUPPCState *env = &cpu->env;
445     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
446     int index = spapr_get_vcpu_id(cpu);
447     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
448                        0xffffffff, 0xffffffff};
449     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
450         : SPAPR_TIMEBASE_FREQ;
451     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
452     uint32_t page_sizes_prop[64];
453     size_t page_sizes_prop_size;
454     uint32_t vcpus_per_socket = smp_threads * smp_cores;
455     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
456     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
457     sPAPRDRConnector *drc;
458     int drc_index;
459     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
460     int i;
461 
462     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
463     if (drc) {
464         drc_index = spapr_drc_index(drc);
465         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
466     }
467 
468     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
469     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
470 
471     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
472     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
473                            env->dcache_line_size)));
474     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
475                            env->dcache_line_size)));
476     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
477                            env->icache_line_size)));
478     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
479                            env->icache_line_size)));
480 
481     if (pcc->l1_dcache_size) {
482         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
483                                pcc->l1_dcache_size)));
484     } else {
485         warn_report("Unknown L1 dcache size for cpu");
486     }
487     if (pcc->l1_icache_size) {
488         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
489                                pcc->l1_icache_size)));
490     } else {
491         warn_report("Unknown L1 icache size for cpu");
492     }
493 
494     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
495     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
496     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
497     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
498     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
499     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
500 
501     if (env->spr_cb[SPR_PURR].oea_read) {
502         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
503     }
504 
505     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
506         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
507                           segs, sizeof(segs))));
508     }
509 
510     /* Advertise VSX (vector extensions) if available
511      *   1               == VMX / Altivec available
512      *   2               == VSX available
513      *
514      * Only CPUs for which we create core types in spapr_cpu_core.c
515      * are possible, and all of those have VMX */
516     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
517         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
518     } else {
519         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
520     }
521 
522     /* Advertise DFP (Decimal Floating Point) if available
523      *   0 / no property == no DFP
524      *   1               == DFP available */
525     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
526         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
527     }
528 
529     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
530                                                       sizeof(page_sizes_prop));
531     if (page_sizes_prop_size) {
532         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
533                           page_sizes_prop, page_sizes_prop_size)));
534     }
535 
536     spapr_populate_pa_features(spapr, cpu, fdt, offset, false);
537 
538     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
539                            cs->cpu_index / vcpus_per_socket)));
540 
541     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
542                       pft_size_prop, sizeof(pft_size_prop))));
543 
544     if (nb_numa_nodes > 1) {
545         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
546     }
547 
548     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
549 
550     if (pcc->radix_page_info) {
551         for (i = 0; i < pcc->radix_page_info->count; i++) {
552             radix_AP_encodings[i] =
553                 cpu_to_be32(pcc->radix_page_info->entries[i]);
554         }
555         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
556                           radix_AP_encodings,
557                           pcc->radix_page_info->count *
558                           sizeof(radix_AP_encodings[0]))));
559     }
560 }
561 
562 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
563 {
564     CPUState **rev;
565     CPUState *cs;
566     int n_cpus;
567     int cpus_offset;
568     char *nodename;
569     int i;
570 
571     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
572     _FDT(cpus_offset);
573     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
574     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
575 
576     /*
577      * We walk the CPUs in reverse order to ensure that CPU DT nodes
578      * created by fdt_add_subnode() end up in the right order in FDT
579      * for the guest kernel the enumerate the CPUs correctly.
580      *
581      * The CPU list cannot be traversed in reverse order, so we need
582      * to do extra work.
583      */
584     n_cpus = 0;
585     rev = NULL;
586     CPU_FOREACH(cs) {
587         rev = g_renew(CPUState *, rev, n_cpus + 1);
588         rev[n_cpus++] = cs;
589     }
590 
591     for (i = n_cpus - 1; i >= 0; i--) {
592         CPUState *cs = rev[i];
593         PowerPCCPU *cpu = POWERPC_CPU(cs);
594         int index = spapr_get_vcpu_id(cpu);
595         DeviceClass *dc = DEVICE_GET_CLASS(cs);
596         int offset;
597 
598         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
599             continue;
600         }
601 
602         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
603         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
604         g_free(nodename);
605         _FDT(offset);
606         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
607     }
608 
609     g_free(rev);
610 }
611 
612 static int spapr_rng_populate_dt(void *fdt)
613 {
614     int node;
615     int ret;
616 
617     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
618     if (node <= 0) {
619         return -1;
620     }
621     ret = fdt_setprop_string(fdt, node, "device_type",
622                              "ibm,platform-facilities");
623     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
624     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
625 
626     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
627     if (node <= 0) {
628         return -1;
629     }
630     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
631 
632     return ret ? -1 : 0;
633 }
634 
635 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
636 {
637     MemoryDeviceInfoList *info;
638 
639     for (info = list; info; info = info->next) {
640         MemoryDeviceInfo *value = info->value;
641 
642         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
643             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
644 
645             if (addr >= pcdimm_info->addr &&
646                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
647                 return pcdimm_info->node;
648             }
649         }
650     }
651 
652     return -1;
653 }
654 
655 struct sPAPRDrconfCellV2 {
656      uint32_t seq_lmbs;
657      uint64_t base_addr;
658      uint32_t drc_index;
659      uint32_t aa_index;
660      uint32_t flags;
661 } QEMU_PACKED;
662 
663 typedef struct DrconfCellQueue {
664     struct sPAPRDrconfCellV2 cell;
665     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
666 } DrconfCellQueue;
667 
668 static DrconfCellQueue *
669 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
670                       uint32_t drc_index, uint32_t aa_index,
671                       uint32_t flags)
672 {
673     DrconfCellQueue *elem;
674 
675     elem = g_malloc0(sizeof(*elem));
676     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
677     elem->cell.base_addr = cpu_to_be64(base_addr);
678     elem->cell.drc_index = cpu_to_be32(drc_index);
679     elem->cell.aa_index = cpu_to_be32(aa_index);
680     elem->cell.flags = cpu_to_be32(flags);
681 
682     return elem;
683 }
684 
685 /* ibm,dynamic-memory-v2 */
686 static int spapr_populate_drmem_v2(sPAPRMachineState *spapr, void *fdt,
687                                    int offset, MemoryDeviceInfoList *dimms)
688 {
689     MachineState *machine = MACHINE(spapr);
690     uint8_t *int_buf, *cur_index, buf_len;
691     int ret;
692     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
693     uint64_t addr, cur_addr, size;
694     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
695     uint64_t mem_end = machine->device_memory->base +
696                        memory_region_size(&machine->device_memory->mr);
697     uint32_t node, nr_entries = 0;
698     sPAPRDRConnector *drc;
699     DrconfCellQueue *elem, *next;
700     MemoryDeviceInfoList *info;
701     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
702         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
703 
704     /* Entry to cover RAM and the gap area */
705     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
706                                  SPAPR_LMB_FLAGS_RESERVED |
707                                  SPAPR_LMB_FLAGS_DRC_INVALID);
708     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
709     nr_entries++;
710 
711     cur_addr = machine->device_memory->base;
712     for (info = dimms; info; info = info->next) {
713         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
714 
715         addr = di->addr;
716         size = di->size;
717         node = di->node;
718 
719         /* Entry for hot-pluggable area */
720         if (cur_addr < addr) {
721             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
722             g_assert(drc);
723             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
724                                          cur_addr, spapr_drc_index(drc), -1, 0);
725             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
726             nr_entries++;
727         }
728 
729         /* Entry for DIMM */
730         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
731         g_assert(drc);
732         elem = spapr_get_drconf_cell(size / lmb_size, addr,
733                                      spapr_drc_index(drc), node,
734                                      SPAPR_LMB_FLAGS_ASSIGNED);
735         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
736         nr_entries++;
737         cur_addr = addr + size;
738     }
739 
740     /* Entry for remaining hotpluggable area */
741     if (cur_addr < mem_end) {
742         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
743         g_assert(drc);
744         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
745                                      cur_addr, spapr_drc_index(drc), -1, 0);
746         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
747         nr_entries++;
748     }
749 
750     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
751     int_buf = cur_index = g_malloc0(buf_len);
752     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
753     cur_index += sizeof(nr_entries);
754 
755     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
756         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
757         cur_index += sizeof(elem->cell);
758         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
759         g_free(elem);
760     }
761 
762     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
763     g_free(int_buf);
764     if (ret < 0) {
765         return -1;
766     }
767     return 0;
768 }
769 
770 /* ibm,dynamic-memory */
771 static int spapr_populate_drmem_v1(sPAPRMachineState *spapr, void *fdt,
772                                    int offset, MemoryDeviceInfoList *dimms)
773 {
774     MachineState *machine = MACHINE(spapr);
775     int i, ret;
776     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
777     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
778     uint32_t nr_lmbs = (machine->device_memory->base +
779                        memory_region_size(&machine->device_memory->mr)) /
780                        lmb_size;
781     uint32_t *int_buf, *cur_index, buf_len;
782 
783     /*
784      * Allocate enough buffer size to fit in ibm,dynamic-memory
785      */
786     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
787     cur_index = int_buf = g_malloc0(buf_len);
788     int_buf[0] = cpu_to_be32(nr_lmbs);
789     cur_index++;
790     for (i = 0; i < nr_lmbs; i++) {
791         uint64_t addr = i * lmb_size;
792         uint32_t *dynamic_memory = cur_index;
793 
794         if (i >= device_lmb_start) {
795             sPAPRDRConnector *drc;
796 
797             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
798             g_assert(drc);
799 
800             dynamic_memory[0] = cpu_to_be32(addr >> 32);
801             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
802             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
803             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
804             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
805             if (memory_region_present(get_system_memory(), addr)) {
806                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
807             } else {
808                 dynamic_memory[5] = cpu_to_be32(0);
809             }
810         } else {
811             /*
812              * LMB information for RMA, boot time RAM and gap b/n RAM and
813              * device memory region -- all these are marked as reserved
814              * and as having no valid DRC.
815              */
816             dynamic_memory[0] = cpu_to_be32(addr >> 32);
817             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
818             dynamic_memory[2] = cpu_to_be32(0);
819             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
820             dynamic_memory[4] = cpu_to_be32(-1);
821             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
822                                             SPAPR_LMB_FLAGS_DRC_INVALID);
823         }
824 
825         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
826     }
827     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
828     g_free(int_buf);
829     if (ret < 0) {
830         return -1;
831     }
832     return 0;
833 }
834 
835 /*
836  * Adds ibm,dynamic-reconfiguration-memory node.
837  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
838  * of this device tree node.
839  */
840 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
841 {
842     MachineState *machine = MACHINE(spapr);
843     int ret, i, offset;
844     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
845     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
846     uint32_t *int_buf, *cur_index, buf_len;
847     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
848     MemoryDeviceInfoList *dimms = NULL;
849 
850     /*
851      * Don't create the node if there is no device memory
852      */
853     if (machine->ram_size == machine->maxram_size) {
854         return 0;
855     }
856 
857     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
858 
859     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
860                     sizeof(prop_lmb_size));
861     if (ret < 0) {
862         return ret;
863     }
864 
865     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
866     if (ret < 0) {
867         return ret;
868     }
869 
870     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
871     if (ret < 0) {
872         return ret;
873     }
874 
875     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
876     dimms = qmp_memory_device_list();
877     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
878         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
879     } else {
880         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
881     }
882     qapi_free_MemoryDeviceInfoList(dimms);
883 
884     if (ret < 0) {
885         return ret;
886     }
887 
888     /* ibm,associativity-lookup-arrays */
889     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
890     cur_index = int_buf = g_malloc0(buf_len);
891     int_buf[0] = cpu_to_be32(nr_nodes);
892     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
893     cur_index += 2;
894     for (i = 0; i < nr_nodes; i++) {
895         uint32_t associativity[] = {
896             cpu_to_be32(0x0),
897             cpu_to_be32(0x0),
898             cpu_to_be32(0x0),
899             cpu_to_be32(i)
900         };
901         memcpy(cur_index, associativity, sizeof(associativity));
902         cur_index += 4;
903     }
904     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
905             (cur_index - int_buf) * sizeof(uint32_t));
906     g_free(int_buf);
907 
908     return ret;
909 }
910 
911 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
912                                 sPAPROptionVector *ov5_updates)
913 {
914     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
915     int ret = 0, offset;
916 
917     /* Generate ibm,dynamic-reconfiguration-memory node if required */
918     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
919         g_assert(smc->dr_lmb_enabled);
920         ret = spapr_populate_drconf_memory(spapr, fdt);
921         if (ret) {
922             goto out;
923         }
924     }
925 
926     offset = fdt_path_offset(fdt, "/chosen");
927     if (offset < 0) {
928         offset = fdt_add_subnode(fdt, 0, "chosen");
929         if (offset < 0) {
930             return offset;
931         }
932     }
933     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
934                                  "ibm,architecture-vec-5");
935 
936 out:
937     return ret;
938 }
939 
940 static bool spapr_hotplugged_dev_before_cas(void)
941 {
942     Object *drc_container, *obj;
943     ObjectProperty *prop;
944     ObjectPropertyIterator iter;
945 
946     drc_container = container_get(object_get_root(), "/dr-connector");
947     object_property_iter_init(&iter, drc_container);
948     while ((prop = object_property_iter_next(&iter))) {
949         if (!strstart(prop->type, "link<", NULL)) {
950             continue;
951         }
952         obj = object_property_get_link(drc_container, prop->name, NULL);
953         if (spapr_drc_needed(obj)) {
954             return true;
955         }
956     }
957     return false;
958 }
959 
960 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
961                                  target_ulong addr, target_ulong size,
962                                  sPAPROptionVector *ov5_updates)
963 {
964     void *fdt, *fdt_skel;
965     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
966 
967     if (spapr_hotplugged_dev_before_cas()) {
968         return 1;
969     }
970 
971     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
972         error_report("SLOF provided an unexpected CAS buffer size "
973                      TARGET_FMT_lu " (min: %zu, max: %u)",
974                      size, sizeof(hdr), FW_MAX_SIZE);
975         exit(EXIT_FAILURE);
976     }
977 
978     size -= sizeof(hdr);
979 
980     /* Create skeleton */
981     fdt_skel = g_malloc0(size);
982     _FDT((fdt_create(fdt_skel, size)));
983     _FDT((fdt_finish_reservemap(fdt_skel)));
984     _FDT((fdt_begin_node(fdt_skel, "")));
985     _FDT((fdt_end_node(fdt_skel)));
986     _FDT((fdt_finish(fdt_skel)));
987     fdt = g_malloc0(size);
988     _FDT((fdt_open_into(fdt_skel, fdt, size)));
989     g_free(fdt_skel);
990 
991     /* Fixup cpu nodes */
992     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
993 
994     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
995         return -1;
996     }
997 
998     /* Pack resulting tree */
999     _FDT((fdt_pack(fdt)));
1000 
1001     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
1002         trace_spapr_cas_failed(size);
1003         return -1;
1004     }
1005 
1006     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
1007     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
1008     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
1009     g_free(fdt);
1010 
1011     return 0;
1012 }
1013 
1014 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
1015 {
1016     int rtas;
1017     GString *hypertas = g_string_sized_new(256);
1018     GString *qemu_hypertas = g_string_sized_new(256);
1019     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1020     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
1021         memory_region_size(&MACHINE(spapr)->device_memory->mr);
1022     uint32_t lrdr_capacity[] = {
1023         cpu_to_be32(max_device_addr >> 32),
1024         cpu_to_be32(max_device_addr & 0xffffffff),
1025         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1026         cpu_to_be32(max_cpus / smp_threads),
1027     };
1028     uint32_t maxdomains[] = {
1029         cpu_to_be32(4),
1030         cpu_to_be32(0),
1031         cpu_to_be32(0),
1032         cpu_to_be32(0),
1033         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes : 1),
1034     };
1035 
1036     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1037 
1038     /* hypertas */
1039     add_str(hypertas, "hcall-pft");
1040     add_str(hypertas, "hcall-term");
1041     add_str(hypertas, "hcall-dabr");
1042     add_str(hypertas, "hcall-interrupt");
1043     add_str(hypertas, "hcall-tce");
1044     add_str(hypertas, "hcall-vio");
1045     add_str(hypertas, "hcall-splpar");
1046     add_str(hypertas, "hcall-bulk");
1047     add_str(hypertas, "hcall-set-mode");
1048     add_str(hypertas, "hcall-sprg0");
1049     add_str(hypertas, "hcall-copy");
1050     add_str(hypertas, "hcall-debug");
1051     add_str(hypertas, "hcall-vphn");
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 {
1230     MachineState *machine = MACHINE(spapr);
1231     MachineClass *mc = MACHINE_GET_CLASS(machine);
1232     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1233     int ret;
1234     void *fdt;
1235     sPAPRPHBState *phb;
1236     char *buf;
1237 
1238     fdt = g_malloc0(FDT_MAX_SIZE);
1239     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1240 
1241     /* Root node */
1242     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1243     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1244     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1245 
1246     /*
1247      * Add info to guest to indentify which host is it being run on
1248      * and what is the uuid of the guest
1249      */
1250     if (kvmppc_get_host_model(&buf)) {
1251         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1252         g_free(buf);
1253     }
1254     if (kvmppc_get_host_serial(&buf)) {
1255         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1256         g_free(buf);
1257     }
1258 
1259     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1260 
1261     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1262     if (qemu_uuid_set) {
1263         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1264     }
1265     g_free(buf);
1266 
1267     if (qemu_get_vm_name()) {
1268         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1269                                 qemu_get_vm_name()));
1270     }
1271 
1272     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1273     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1274 
1275     /* /interrupt controller */
1276     spapr->irq->dt_populate(spapr, spapr_max_server_number(spapr), fdt,
1277                           PHANDLE_XICP);
1278 
1279     ret = spapr_populate_memory(spapr, fdt);
1280     if (ret < 0) {
1281         error_report("couldn't setup memory nodes in fdt");
1282         exit(1);
1283     }
1284 
1285     /* /vdevice */
1286     spapr_dt_vdevice(spapr->vio_bus, fdt);
1287 
1288     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1289         ret = spapr_rng_populate_dt(fdt);
1290         if (ret < 0) {
1291             error_report("could not set up rng device in the fdt");
1292             exit(1);
1293         }
1294     }
1295 
1296     QLIST_FOREACH(phb, &spapr->phbs, list) {
1297         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt,
1298                                     spapr->irq->nr_msis);
1299         if (ret < 0) {
1300             error_report("couldn't setup PCI devices in fdt");
1301             exit(1);
1302         }
1303     }
1304 
1305     /* cpus */
1306     spapr_populate_cpus_dt_node(fdt, spapr);
1307 
1308     if (smc->dr_lmb_enabled) {
1309         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1310     }
1311 
1312     if (mc->has_hotpluggable_cpus) {
1313         int offset = fdt_path_offset(fdt, "/cpus");
1314         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1315                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1316         if (ret < 0) {
1317             error_report("Couldn't set up CPU DR device tree properties");
1318             exit(1);
1319         }
1320     }
1321 
1322     /* /event-sources */
1323     spapr_dt_events(spapr, fdt);
1324 
1325     /* /rtas */
1326     spapr_dt_rtas(spapr, fdt);
1327 
1328     /* /chosen */
1329     spapr_dt_chosen(spapr, fdt);
1330 
1331     /* /hypervisor */
1332     if (kvm_enabled()) {
1333         spapr_dt_hypervisor(spapr, fdt);
1334     }
1335 
1336     /* Build memory reserve map */
1337     if (spapr->kernel_size) {
1338         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1339     }
1340     if (spapr->initrd_size) {
1341         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1342     }
1343 
1344     /* ibm,client-architecture-support updates */
1345     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1346     if (ret < 0) {
1347         error_report("couldn't setup CAS properties fdt");
1348         exit(1);
1349     }
1350 
1351     return fdt;
1352 }
1353 
1354 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1355 {
1356     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1357 }
1358 
1359 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1360                                     PowerPCCPU *cpu)
1361 {
1362     CPUPPCState *env = &cpu->env;
1363 
1364     /* The TCG path should also be holding the BQL at this point */
1365     g_assert(qemu_mutex_iothread_locked());
1366 
1367     if (msr_pr) {
1368         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1369         env->gpr[3] = H_PRIVILEGE;
1370     } else {
1371         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1372     }
1373 }
1374 
1375 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1376 {
1377     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1378 
1379     return spapr->patb_entry;
1380 }
1381 
1382 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1383 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1384 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1385 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1386 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1387 
1388 /*
1389  * Get the fd to access the kernel htab, re-opening it if necessary
1390  */
1391 static int get_htab_fd(sPAPRMachineState *spapr)
1392 {
1393     Error *local_err = NULL;
1394 
1395     if (spapr->htab_fd >= 0) {
1396         return spapr->htab_fd;
1397     }
1398 
1399     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1400     if (spapr->htab_fd < 0) {
1401         error_report_err(local_err);
1402     }
1403 
1404     return spapr->htab_fd;
1405 }
1406 
1407 void close_htab_fd(sPAPRMachineState *spapr)
1408 {
1409     if (spapr->htab_fd >= 0) {
1410         close(spapr->htab_fd);
1411     }
1412     spapr->htab_fd = -1;
1413 }
1414 
1415 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1416 {
1417     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1418 
1419     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1420 }
1421 
1422 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1423 {
1424     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1425 
1426     assert(kvm_enabled());
1427 
1428     if (!spapr->htab) {
1429         return 0;
1430     }
1431 
1432     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1433 }
1434 
1435 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1436                                                 hwaddr ptex, int n)
1437 {
1438     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1439     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1440 
1441     if (!spapr->htab) {
1442         /*
1443          * HTAB is controlled by KVM. Fetch into temporary buffer
1444          */
1445         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1446         kvmppc_read_hptes(hptes, ptex, n);
1447         return hptes;
1448     }
1449 
1450     /*
1451      * HTAB is controlled by QEMU. Just point to the internally
1452      * accessible PTEG.
1453      */
1454     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1455 }
1456 
1457 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1458                               const ppc_hash_pte64_t *hptes,
1459                               hwaddr ptex, int n)
1460 {
1461     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1462 
1463     if (!spapr->htab) {
1464         g_free((void *)hptes);
1465     }
1466 
1467     /* Nothing to do for qemu managed HPT */
1468 }
1469 
1470 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1471                              uint64_t pte0, uint64_t pte1)
1472 {
1473     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1474     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1475 
1476     if (!spapr->htab) {
1477         kvmppc_write_hpte(ptex, pte0, pte1);
1478     } else {
1479         stq_p(spapr->htab + offset, pte0);
1480         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1481     }
1482 }
1483 
1484 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1485 {
1486     int shift;
1487 
1488     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1489      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1490      * that's much more than is needed for Linux guests */
1491     shift = ctz64(pow2ceil(ramsize)) - 7;
1492     shift = MAX(shift, 18); /* Minimum architected size */
1493     shift = MIN(shift, 46); /* Maximum architected size */
1494     return shift;
1495 }
1496 
1497 void spapr_free_hpt(sPAPRMachineState *spapr)
1498 {
1499     g_free(spapr->htab);
1500     spapr->htab = NULL;
1501     spapr->htab_shift = 0;
1502     close_htab_fd(spapr);
1503 }
1504 
1505 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1506                           Error **errp)
1507 {
1508     long rc;
1509 
1510     /* Clean up any HPT info from a previous boot */
1511     spapr_free_hpt(spapr);
1512 
1513     rc = kvmppc_reset_htab(shift);
1514     if (rc < 0) {
1515         /* kernel-side HPT needed, but couldn't allocate one */
1516         error_setg_errno(errp, errno,
1517                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1518                          shift);
1519         /* This is almost certainly fatal, but if the caller really
1520          * wants to carry on with shift == 0, it's welcome to try */
1521     } else if (rc > 0) {
1522         /* kernel-side HPT allocated */
1523         if (rc != shift) {
1524             error_setg(errp,
1525                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1526                        shift, rc);
1527         }
1528 
1529         spapr->htab_shift = shift;
1530         spapr->htab = NULL;
1531     } else {
1532         /* kernel-side HPT not needed, allocate in userspace instead */
1533         size_t size = 1ULL << shift;
1534         int i;
1535 
1536         spapr->htab = qemu_memalign(size, size);
1537         if (!spapr->htab) {
1538             error_setg_errno(errp, errno,
1539                              "Could not allocate HPT of order %d", shift);
1540             return;
1541         }
1542 
1543         memset(spapr->htab, 0, size);
1544         spapr->htab_shift = shift;
1545 
1546         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1547             DIRTY_HPTE(HPTE(spapr->htab, i));
1548         }
1549     }
1550     /* We're setting up a hash table, so that means we're not radix */
1551     spapr->patb_entry = 0;
1552 }
1553 
1554 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1555 {
1556     int hpt_shift;
1557 
1558     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1559         || (spapr->cas_reboot
1560             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1561         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1562     } else {
1563         uint64_t current_ram_size;
1564 
1565         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1566         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1567     }
1568     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1569 
1570     if (spapr->vrma_adjust) {
1571         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1572                                           spapr->htab_shift);
1573     }
1574 }
1575 
1576 static int spapr_reset_drcs(Object *child, void *opaque)
1577 {
1578     sPAPRDRConnector *drc =
1579         (sPAPRDRConnector *) object_dynamic_cast(child,
1580                                                  TYPE_SPAPR_DR_CONNECTOR);
1581 
1582     if (drc) {
1583         spapr_drc_reset(drc);
1584     }
1585 
1586     return 0;
1587 }
1588 
1589 static void spapr_machine_reset(void)
1590 {
1591     MachineState *machine = MACHINE(qdev_get_machine());
1592     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1593     PowerPCCPU *first_ppc_cpu;
1594     uint32_t rtas_limit;
1595     hwaddr rtas_addr, fdt_addr;
1596     void *fdt;
1597     int rc;
1598 
1599     spapr_caps_apply(spapr);
1600 
1601     first_ppc_cpu = POWERPC_CPU(first_cpu);
1602     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1603         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1604                               spapr->max_compat_pvr)) {
1605         /* If using KVM with radix mode available, VCPUs can be started
1606          * without a HPT because KVM will start them in radix mode.
1607          * Set the GR bit in PATB so that we know there is no HPT. */
1608         spapr->patb_entry = PATBE1_GR;
1609     } else {
1610         spapr_setup_hpt_and_vrma(spapr);
1611     }
1612 
1613     /* if this reset wasn't generated by CAS, we should reset our
1614      * negotiated options and start from scratch */
1615     if (!spapr->cas_reboot) {
1616         spapr_ovec_cleanup(spapr->ov5_cas);
1617         spapr->ov5_cas = spapr_ovec_new();
1618 
1619         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1620     }
1621 
1622     if (!SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
1623         spapr_irq_msi_reset(spapr);
1624     }
1625 
1626     qemu_devices_reset();
1627 
1628     /*
1629      * This is fixing some of the default configuration of the XIVE
1630      * devices. To be called after the reset of the machine devices.
1631      */
1632     spapr_irq_reset(spapr, &error_fatal);
1633 
1634     /* DRC reset may cause a device to be unplugged. This will cause troubles
1635      * if this device is used by another device (eg, a running vhost backend
1636      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1637      * situations, we reset DRCs after all devices have been reset.
1638      */
1639     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1640 
1641     spapr_clear_pending_events(spapr);
1642 
1643     /*
1644      * We place the device tree and RTAS just below either the top of the RMA,
1645      * or just below 2GB, whichever is lower, so that it can be
1646      * processed with 32-bit real mode code if necessary
1647      */
1648     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1649     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1650     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1651 
1652     fdt = spapr_build_fdt(spapr);
1653 
1654     spapr_load_rtas(spapr, fdt, rtas_addr);
1655 
1656     rc = fdt_pack(fdt);
1657 
1658     /* Should only fail if we've built a corrupted tree */
1659     assert(rc == 0);
1660 
1661     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1662         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1663                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1664         exit(1);
1665     }
1666 
1667     /* Load the fdt */
1668     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1669     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1670     g_free(spapr->fdt_blob);
1671     spapr->fdt_size = fdt_totalsize(fdt);
1672     spapr->fdt_initial_size = spapr->fdt_size;
1673     spapr->fdt_blob = fdt;
1674 
1675     /* Set up the entry state */
1676     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1677     first_ppc_cpu->env.gpr[5] = 0;
1678 
1679     spapr->cas_reboot = false;
1680 }
1681 
1682 static void spapr_create_nvram(sPAPRMachineState *spapr)
1683 {
1684     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1685     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1686 
1687     if (dinfo) {
1688         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1689                             &error_fatal);
1690     }
1691 
1692     qdev_init_nofail(dev);
1693 
1694     spapr->nvram = (struct sPAPRNVRAM *)dev;
1695 }
1696 
1697 static void spapr_rtc_create(sPAPRMachineState *spapr)
1698 {
1699     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1700     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1701                               &error_fatal);
1702     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1703                               &error_fatal);
1704     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1705                               "date", &error_fatal);
1706 }
1707 
1708 /* Returns whether we want to use VGA or not */
1709 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1710 {
1711     switch (vga_interface_type) {
1712     case VGA_NONE:
1713         return false;
1714     case VGA_DEVICE:
1715         return true;
1716     case VGA_STD:
1717     case VGA_VIRTIO:
1718     case VGA_CIRRUS:
1719         return pci_vga_init(pci_bus) != NULL;
1720     default:
1721         error_setg(errp,
1722                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1723         return false;
1724     }
1725 }
1726 
1727 static int spapr_pre_load(void *opaque)
1728 {
1729     int rc;
1730 
1731     rc = spapr_caps_pre_load(opaque);
1732     if (rc) {
1733         return rc;
1734     }
1735 
1736     return 0;
1737 }
1738 
1739 static int spapr_post_load(void *opaque, int version_id)
1740 {
1741     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1742     int err = 0;
1743 
1744     err = spapr_caps_post_migration(spapr);
1745     if (err) {
1746         return err;
1747     }
1748 
1749     /*
1750      * In earlier versions, there was no separate qdev for the PAPR
1751      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1752      * So when migrating from those versions, poke the incoming offset
1753      * value into the RTC device
1754      */
1755     if (version_id < 3) {
1756         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1757         if (err) {
1758             return err;
1759         }
1760     }
1761 
1762     if (kvm_enabled() && spapr->patb_entry) {
1763         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1764         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1765         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1766 
1767         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1768         if (err) {
1769             error_report("Process table config unsupported by the host");
1770             return -EINVAL;
1771         }
1772     }
1773 
1774     err = spapr_irq_post_load(spapr, version_id);
1775     if (err) {
1776         return err;
1777     }
1778 
1779     return err;
1780 }
1781 
1782 static int spapr_pre_save(void *opaque)
1783 {
1784     int rc;
1785 
1786     rc = spapr_caps_pre_save(opaque);
1787     if (rc) {
1788         return rc;
1789     }
1790 
1791     return 0;
1792 }
1793 
1794 static bool version_before_3(void *opaque, int version_id)
1795 {
1796     return version_id < 3;
1797 }
1798 
1799 static bool spapr_pending_events_needed(void *opaque)
1800 {
1801     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1802     return !QTAILQ_EMPTY(&spapr->pending_events);
1803 }
1804 
1805 static const VMStateDescription vmstate_spapr_event_entry = {
1806     .name = "spapr_event_log_entry",
1807     .version_id = 1,
1808     .minimum_version_id = 1,
1809     .fields = (VMStateField[]) {
1810         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1811         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1812         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1813                                      NULL, extended_length),
1814         VMSTATE_END_OF_LIST()
1815     },
1816 };
1817 
1818 static const VMStateDescription vmstate_spapr_pending_events = {
1819     .name = "spapr_pending_events",
1820     .version_id = 1,
1821     .minimum_version_id = 1,
1822     .needed = spapr_pending_events_needed,
1823     .fields = (VMStateField[]) {
1824         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1825                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1826         VMSTATE_END_OF_LIST()
1827     },
1828 };
1829 
1830 static bool spapr_ov5_cas_needed(void *opaque)
1831 {
1832     sPAPRMachineState *spapr = opaque;
1833     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1834     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1835     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1836     bool cas_needed;
1837 
1838     /* Prior to the introduction of sPAPROptionVector, we had two option
1839      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1840      * Both of these options encode machine topology into the device-tree
1841      * in such a way that the now-booted OS should still be able to interact
1842      * appropriately with QEMU regardless of what options were actually
1843      * negotiatied on the source side.
1844      *
1845      * As such, we can avoid migrating the CAS-negotiated options if these
1846      * are the only options available on the current machine/platform.
1847      * Since these are the only options available for pseries-2.7 and
1848      * earlier, this allows us to maintain old->new/new->old migration
1849      * compatibility.
1850      *
1851      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1852      * via default pseries-2.8 machines and explicit command-line parameters.
1853      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1854      * of the actual CAS-negotiated values to continue working properly. For
1855      * example, availability of memory unplug depends on knowing whether
1856      * OV5_HP_EVT was negotiated via CAS.
1857      *
1858      * Thus, for any cases where the set of available CAS-negotiatable
1859      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1860      * include the CAS-negotiated options in the migration stream, unless
1861      * if they affect boot time behaviour only.
1862      */
1863     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1864     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1865     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1866 
1867     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1868      * the mask itself since in the future it's possible "legacy" bits may be
1869      * removed via machine options, which could generate a false positive
1870      * that breaks migration.
1871      */
1872     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1873     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1874 
1875     spapr_ovec_cleanup(ov5_mask);
1876     spapr_ovec_cleanup(ov5_legacy);
1877     spapr_ovec_cleanup(ov5_removed);
1878 
1879     return cas_needed;
1880 }
1881 
1882 static const VMStateDescription vmstate_spapr_ov5_cas = {
1883     .name = "spapr_option_vector_ov5_cas",
1884     .version_id = 1,
1885     .minimum_version_id = 1,
1886     .needed = spapr_ov5_cas_needed,
1887     .fields = (VMStateField[]) {
1888         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1889                                  vmstate_spapr_ovec, sPAPROptionVector),
1890         VMSTATE_END_OF_LIST()
1891     },
1892 };
1893 
1894 static bool spapr_patb_entry_needed(void *opaque)
1895 {
1896     sPAPRMachineState *spapr = opaque;
1897 
1898     return !!spapr->patb_entry;
1899 }
1900 
1901 static const VMStateDescription vmstate_spapr_patb_entry = {
1902     .name = "spapr_patb_entry",
1903     .version_id = 1,
1904     .minimum_version_id = 1,
1905     .needed = spapr_patb_entry_needed,
1906     .fields = (VMStateField[]) {
1907         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1908         VMSTATE_END_OF_LIST()
1909     },
1910 };
1911 
1912 static bool spapr_irq_map_needed(void *opaque)
1913 {
1914     sPAPRMachineState *spapr = opaque;
1915 
1916     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1917 }
1918 
1919 static const VMStateDescription vmstate_spapr_irq_map = {
1920     .name = "spapr_irq_map",
1921     .version_id = 1,
1922     .minimum_version_id = 1,
1923     .needed = spapr_irq_map_needed,
1924     .fields = (VMStateField[]) {
1925         VMSTATE_BITMAP(irq_map, sPAPRMachineState, 0, irq_map_nr),
1926         VMSTATE_END_OF_LIST()
1927     },
1928 };
1929 
1930 static bool spapr_dtb_needed(void *opaque)
1931 {
1932     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1933 
1934     return smc->update_dt_enabled;
1935 }
1936 
1937 static int spapr_dtb_pre_load(void *opaque)
1938 {
1939     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1940 
1941     g_free(spapr->fdt_blob);
1942     spapr->fdt_blob = NULL;
1943     spapr->fdt_size = 0;
1944 
1945     return 0;
1946 }
1947 
1948 static const VMStateDescription vmstate_spapr_dtb = {
1949     .name = "spapr_dtb",
1950     .version_id = 1,
1951     .minimum_version_id = 1,
1952     .needed = spapr_dtb_needed,
1953     .pre_load = spapr_dtb_pre_load,
1954     .fields = (VMStateField[]) {
1955         VMSTATE_UINT32(fdt_initial_size, sPAPRMachineState),
1956         VMSTATE_UINT32(fdt_size, sPAPRMachineState),
1957         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, sPAPRMachineState, 0, NULL,
1958                                      fdt_size),
1959         VMSTATE_END_OF_LIST()
1960     },
1961 };
1962 
1963 static const VMStateDescription vmstate_spapr = {
1964     .name = "spapr",
1965     .version_id = 3,
1966     .minimum_version_id = 1,
1967     .pre_load = spapr_pre_load,
1968     .post_load = spapr_post_load,
1969     .pre_save = spapr_pre_save,
1970     .fields = (VMStateField[]) {
1971         /* used to be @next_irq */
1972         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1973 
1974         /* RTC offset */
1975         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1976 
1977         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1978         VMSTATE_END_OF_LIST()
1979     },
1980     .subsections = (const VMStateDescription*[]) {
1981         &vmstate_spapr_ov5_cas,
1982         &vmstate_spapr_patb_entry,
1983         &vmstate_spapr_pending_events,
1984         &vmstate_spapr_cap_htm,
1985         &vmstate_spapr_cap_vsx,
1986         &vmstate_spapr_cap_dfp,
1987         &vmstate_spapr_cap_cfpc,
1988         &vmstate_spapr_cap_sbbc,
1989         &vmstate_spapr_cap_ibs,
1990         &vmstate_spapr_irq_map,
1991         &vmstate_spapr_cap_nested_kvm_hv,
1992         &vmstate_spapr_dtb,
1993         NULL
1994     }
1995 };
1996 
1997 static int htab_save_setup(QEMUFile *f, void *opaque)
1998 {
1999     sPAPRMachineState *spapr = opaque;
2000 
2001     /* "Iteration" header */
2002     if (!spapr->htab_shift) {
2003         qemu_put_be32(f, -1);
2004     } else {
2005         qemu_put_be32(f, spapr->htab_shift);
2006     }
2007 
2008     if (spapr->htab) {
2009         spapr->htab_save_index = 0;
2010         spapr->htab_first_pass = true;
2011     } else {
2012         if (spapr->htab_shift) {
2013             assert(kvm_enabled());
2014         }
2015     }
2016 
2017 
2018     return 0;
2019 }
2020 
2021 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
2022                             int chunkstart, int n_valid, int n_invalid)
2023 {
2024     qemu_put_be32(f, chunkstart);
2025     qemu_put_be16(f, n_valid);
2026     qemu_put_be16(f, n_invalid);
2027     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2028                     HASH_PTE_SIZE_64 * n_valid);
2029 }
2030 
2031 static void htab_save_end_marker(QEMUFile *f)
2032 {
2033     qemu_put_be32(f, 0);
2034     qemu_put_be16(f, 0);
2035     qemu_put_be16(f, 0);
2036 }
2037 
2038 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
2039                                  int64_t max_ns)
2040 {
2041     bool has_timeout = max_ns != -1;
2042     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2043     int index = spapr->htab_save_index;
2044     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2045 
2046     assert(spapr->htab_first_pass);
2047 
2048     do {
2049         int chunkstart;
2050 
2051         /* Consume invalid HPTEs */
2052         while ((index < htabslots)
2053                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2054             CLEAN_HPTE(HPTE(spapr->htab, index));
2055             index++;
2056         }
2057 
2058         /* Consume valid HPTEs */
2059         chunkstart = index;
2060         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2061                && HPTE_VALID(HPTE(spapr->htab, index))) {
2062             CLEAN_HPTE(HPTE(spapr->htab, index));
2063             index++;
2064         }
2065 
2066         if (index > chunkstart) {
2067             int n_valid = index - chunkstart;
2068 
2069             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2070 
2071             if (has_timeout &&
2072                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2073                 break;
2074             }
2075         }
2076     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2077 
2078     if (index >= htabslots) {
2079         assert(index == htabslots);
2080         index = 0;
2081         spapr->htab_first_pass = false;
2082     }
2083     spapr->htab_save_index = index;
2084 }
2085 
2086 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
2087                                 int64_t max_ns)
2088 {
2089     bool final = max_ns < 0;
2090     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2091     int examined = 0, sent = 0;
2092     int index = spapr->htab_save_index;
2093     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2094 
2095     assert(!spapr->htab_first_pass);
2096 
2097     do {
2098         int chunkstart, invalidstart;
2099 
2100         /* Consume non-dirty HPTEs */
2101         while ((index < htabslots)
2102                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2103             index++;
2104             examined++;
2105         }
2106 
2107         chunkstart = index;
2108         /* Consume valid dirty HPTEs */
2109         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2110                && HPTE_DIRTY(HPTE(spapr->htab, index))
2111                && HPTE_VALID(HPTE(spapr->htab, index))) {
2112             CLEAN_HPTE(HPTE(spapr->htab, index));
2113             index++;
2114             examined++;
2115         }
2116 
2117         invalidstart = index;
2118         /* Consume invalid dirty HPTEs */
2119         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2120                && HPTE_DIRTY(HPTE(spapr->htab, index))
2121                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2122             CLEAN_HPTE(HPTE(spapr->htab, index));
2123             index++;
2124             examined++;
2125         }
2126 
2127         if (index > chunkstart) {
2128             int n_valid = invalidstart - chunkstart;
2129             int n_invalid = index - invalidstart;
2130 
2131             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2132             sent += index - chunkstart;
2133 
2134             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2135                 break;
2136             }
2137         }
2138 
2139         if (examined >= htabslots) {
2140             break;
2141         }
2142 
2143         if (index >= htabslots) {
2144             assert(index == htabslots);
2145             index = 0;
2146         }
2147     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2148 
2149     if (index >= htabslots) {
2150         assert(index == htabslots);
2151         index = 0;
2152     }
2153 
2154     spapr->htab_save_index = index;
2155 
2156     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2157 }
2158 
2159 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2160 #define MAX_KVM_BUF_SIZE    2048
2161 
2162 static int htab_save_iterate(QEMUFile *f, void *opaque)
2163 {
2164     sPAPRMachineState *spapr = opaque;
2165     int fd;
2166     int rc = 0;
2167 
2168     /* Iteration header */
2169     if (!spapr->htab_shift) {
2170         qemu_put_be32(f, -1);
2171         return 1;
2172     } else {
2173         qemu_put_be32(f, 0);
2174     }
2175 
2176     if (!spapr->htab) {
2177         assert(kvm_enabled());
2178 
2179         fd = get_htab_fd(spapr);
2180         if (fd < 0) {
2181             return fd;
2182         }
2183 
2184         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2185         if (rc < 0) {
2186             return rc;
2187         }
2188     } else  if (spapr->htab_first_pass) {
2189         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2190     } else {
2191         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2192     }
2193 
2194     htab_save_end_marker(f);
2195 
2196     return rc;
2197 }
2198 
2199 static int htab_save_complete(QEMUFile *f, void *opaque)
2200 {
2201     sPAPRMachineState *spapr = opaque;
2202     int fd;
2203 
2204     /* Iteration header */
2205     if (!spapr->htab_shift) {
2206         qemu_put_be32(f, -1);
2207         return 0;
2208     } else {
2209         qemu_put_be32(f, 0);
2210     }
2211 
2212     if (!spapr->htab) {
2213         int rc;
2214 
2215         assert(kvm_enabled());
2216 
2217         fd = get_htab_fd(spapr);
2218         if (fd < 0) {
2219             return fd;
2220         }
2221 
2222         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2223         if (rc < 0) {
2224             return rc;
2225         }
2226     } else {
2227         if (spapr->htab_first_pass) {
2228             htab_save_first_pass(f, spapr, -1);
2229         }
2230         htab_save_later_pass(f, spapr, -1);
2231     }
2232 
2233     /* End marker */
2234     htab_save_end_marker(f);
2235 
2236     return 0;
2237 }
2238 
2239 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2240 {
2241     sPAPRMachineState *spapr = opaque;
2242     uint32_t section_hdr;
2243     int fd = -1;
2244     Error *local_err = NULL;
2245 
2246     if (version_id < 1 || version_id > 1) {
2247         error_report("htab_load() bad version");
2248         return -EINVAL;
2249     }
2250 
2251     section_hdr = qemu_get_be32(f);
2252 
2253     if (section_hdr == -1) {
2254         spapr_free_hpt(spapr);
2255         return 0;
2256     }
2257 
2258     if (section_hdr) {
2259         /* First section gives the htab size */
2260         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2261         if (local_err) {
2262             error_report_err(local_err);
2263             return -EINVAL;
2264         }
2265         return 0;
2266     }
2267 
2268     if (!spapr->htab) {
2269         assert(kvm_enabled());
2270 
2271         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2272         if (fd < 0) {
2273             error_report_err(local_err);
2274             return fd;
2275         }
2276     }
2277 
2278     while (true) {
2279         uint32_t index;
2280         uint16_t n_valid, n_invalid;
2281 
2282         index = qemu_get_be32(f);
2283         n_valid = qemu_get_be16(f);
2284         n_invalid = qemu_get_be16(f);
2285 
2286         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2287             /* End of Stream */
2288             break;
2289         }
2290 
2291         if ((index + n_valid + n_invalid) >
2292             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2293             /* Bad index in stream */
2294             error_report(
2295                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2296                 index, n_valid, n_invalid, spapr->htab_shift);
2297             return -EINVAL;
2298         }
2299 
2300         if (spapr->htab) {
2301             if (n_valid) {
2302                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2303                                 HASH_PTE_SIZE_64 * n_valid);
2304             }
2305             if (n_invalid) {
2306                 memset(HPTE(spapr->htab, index + n_valid), 0,
2307                        HASH_PTE_SIZE_64 * n_invalid);
2308             }
2309         } else {
2310             int rc;
2311 
2312             assert(fd >= 0);
2313 
2314             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2315             if (rc < 0) {
2316                 return rc;
2317             }
2318         }
2319     }
2320 
2321     if (!spapr->htab) {
2322         assert(fd >= 0);
2323         close(fd);
2324     }
2325 
2326     return 0;
2327 }
2328 
2329 static void htab_save_cleanup(void *opaque)
2330 {
2331     sPAPRMachineState *spapr = opaque;
2332 
2333     close_htab_fd(spapr);
2334 }
2335 
2336 static SaveVMHandlers savevm_htab_handlers = {
2337     .save_setup = htab_save_setup,
2338     .save_live_iterate = htab_save_iterate,
2339     .save_live_complete_precopy = htab_save_complete,
2340     .save_cleanup = htab_save_cleanup,
2341     .load_state = htab_load,
2342 };
2343 
2344 static void spapr_boot_set(void *opaque, const char *boot_device,
2345                            Error **errp)
2346 {
2347     MachineState *machine = MACHINE(opaque);
2348     machine->boot_order = g_strdup(boot_device);
2349 }
2350 
2351 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2352 {
2353     MachineState *machine = MACHINE(spapr);
2354     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2355     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2356     int i;
2357 
2358     for (i = 0; i < nr_lmbs; i++) {
2359         uint64_t addr;
2360 
2361         addr = i * lmb_size + machine->device_memory->base;
2362         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2363                                addr / lmb_size);
2364     }
2365 }
2366 
2367 /*
2368  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2369  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2370  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2371  */
2372 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2373 {
2374     int i;
2375 
2376     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2377         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2378                    " is not aligned to %" PRIu64 " MiB",
2379                    machine->ram_size,
2380                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2381         return;
2382     }
2383 
2384     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2385         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2386                    " is not aligned to %" PRIu64 " MiB",
2387                    machine->ram_size,
2388                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2389         return;
2390     }
2391 
2392     for (i = 0; i < nb_numa_nodes; i++) {
2393         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2394             error_setg(errp,
2395                        "Node %d memory size 0x%" PRIx64
2396                        " is not aligned to %" PRIu64 " MiB",
2397                        i, numa_info[i].node_mem,
2398                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2399             return;
2400         }
2401     }
2402 }
2403 
2404 /* find cpu slot in machine->possible_cpus by core_id */
2405 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2406 {
2407     int index = id / smp_threads;
2408 
2409     if (index >= ms->possible_cpus->len) {
2410         return NULL;
2411     }
2412     if (idx) {
2413         *idx = index;
2414     }
2415     return &ms->possible_cpus->cpus[index];
2416 }
2417 
2418 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2419 {
2420     Error *local_err = NULL;
2421     bool vsmt_user = !!spapr->vsmt;
2422     int kvm_smt = kvmppc_smt_threads();
2423     int ret;
2424 
2425     if (!kvm_enabled() && (smp_threads > 1)) {
2426         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2427                      "on a pseries machine");
2428         goto out;
2429     }
2430     if (!is_power_of_2(smp_threads)) {
2431         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2432                      "machine because it must be a power of 2", smp_threads);
2433         goto out;
2434     }
2435 
2436     /* Detemine the VSMT mode to use: */
2437     if (vsmt_user) {
2438         if (spapr->vsmt < smp_threads) {
2439             error_setg(&local_err, "Cannot support VSMT mode %d"
2440                          " because it must be >= threads/core (%d)",
2441                          spapr->vsmt, smp_threads);
2442             goto out;
2443         }
2444         /* In this case, spapr->vsmt has been set by the command line */
2445     } else {
2446         /*
2447          * Default VSMT value is tricky, because we need it to be as
2448          * consistent as possible (for migration), but this requires
2449          * changing it for at least some existing cases.  We pick 8 as
2450          * the value that we'd get with KVM on POWER8, the
2451          * overwhelmingly common case in production systems.
2452          */
2453         spapr->vsmt = MAX(8, smp_threads);
2454     }
2455 
2456     /* KVM: If necessary, set the SMT mode: */
2457     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2458         ret = kvmppc_set_smt_threads(spapr->vsmt);
2459         if (ret) {
2460             /* Looks like KVM isn't able to change VSMT mode */
2461             error_setg(&local_err,
2462                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2463                        spapr->vsmt, ret);
2464             /* We can live with that if the default one is big enough
2465              * for the number of threads, and a submultiple of the one
2466              * we want.  In this case we'll waste some vcpu ids, but
2467              * behaviour will be correct */
2468             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2469                 warn_report_err(local_err);
2470                 local_err = NULL;
2471                 goto out;
2472             } else {
2473                 if (!vsmt_user) {
2474                     error_append_hint(&local_err,
2475                                       "On PPC, a VM with %d threads/core"
2476                                       " on a host with %d threads/core"
2477                                       " requires the use of VSMT mode %d.\n",
2478                                       smp_threads, kvm_smt, spapr->vsmt);
2479                 }
2480                 kvmppc_hint_smt_possible(&local_err);
2481                 goto out;
2482             }
2483         }
2484     }
2485     /* else TCG: nothing to do currently */
2486 out:
2487     error_propagate(errp, local_err);
2488 }
2489 
2490 static void spapr_init_cpus(sPAPRMachineState *spapr)
2491 {
2492     MachineState *machine = MACHINE(spapr);
2493     MachineClass *mc = MACHINE_GET_CLASS(machine);
2494     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2495     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2496     const CPUArchIdList *possible_cpus;
2497     int boot_cores_nr = smp_cpus / smp_threads;
2498     int i;
2499 
2500     possible_cpus = mc->possible_cpu_arch_ids(machine);
2501     if (mc->has_hotpluggable_cpus) {
2502         if (smp_cpus % smp_threads) {
2503             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2504                          smp_cpus, smp_threads);
2505             exit(1);
2506         }
2507         if (max_cpus % smp_threads) {
2508             error_report("max_cpus (%u) must be multiple of threads (%u)",
2509                          max_cpus, smp_threads);
2510             exit(1);
2511         }
2512     } else {
2513         if (max_cpus != smp_cpus) {
2514             error_report("This machine version does not support CPU hotplug");
2515             exit(1);
2516         }
2517         boot_cores_nr = possible_cpus->len;
2518     }
2519 
2520     if (smc->pre_2_10_has_unused_icps) {
2521         int i;
2522 
2523         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2524             /* Dummy entries get deregistered when real ICPState objects
2525              * are registered during CPU core hotplug.
2526              */
2527             pre_2_10_vmstate_register_dummy_icp(i);
2528         }
2529     }
2530 
2531     for (i = 0; i < possible_cpus->len; i++) {
2532         int core_id = i * smp_threads;
2533 
2534         if (mc->has_hotpluggable_cpus) {
2535             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2536                                    spapr_vcpu_id(spapr, core_id));
2537         }
2538 
2539         if (i < boot_cores_nr) {
2540             Object *core  = object_new(type);
2541             int nr_threads = smp_threads;
2542 
2543             /* Handle the partially filled core for older machine types */
2544             if ((i + 1) * smp_threads >= smp_cpus) {
2545                 nr_threads = smp_cpus - i * smp_threads;
2546             }
2547 
2548             object_property_set_int(core, nr_threads, "nr-threads",
2549                                     &error_fatal);
2550             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2551                                     &error_fatal);
2552             object_property_set_bool(core, true, "realized", &error_fatal);
2553 
2554             object_unref(core);
2555         }
2556     }
2557 }
2558 
2559 static PCIHostState *spapr_create_default_phb(void)
2560 {
2561     DeviceState *dev;
2562 
2563     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2564     qdev_prop_set_uint32(dev, "index", 0);
2565     qdev_init_nofail(dev);
2566 
2567     return PCI_HOST_BRIDGE(dev);
2568 }
2569 
2570 /* pSeries LPAR / sPAPR hardware init */
2571 static void spapr_machine_init(MachineState *machine)
2572 {
2573     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2574     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2575     const char *kernel_filename = machine->kernel_filename;
2576     const char *initrd_filename = machine->initrd_filename;
2577     PCIHostState *phb;
2578     int i;
2579     MemoryRegion *sysmem = get_system_memory();
2580     MemoryRegion *ram = g_new(MemoryRegion, 1);
2581     hwaddr node0_size = spapr_node0_size(machine);
2582     long load_limit, fw_size;
2583     char *filename;
2584     Error *resize_hpt_err = NULL;
2585 
2586     msi_nonbroken = true;
2587 
2588     QLIST_INIT(&spapr->phbs);
2589     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2590 
2591     /* Determine capabilities to run with */
2592     spapr_caps_init(spapr);
2593 
2594     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2595     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2596         /*
2597          * If the user explicitly requested a mode we should either
2598          * supply it, or fail completely (which we do below).  But if
2599          * it's not set explicitly, we reset our mode to something
2600          * that works
2601          */
2602         if (resize_hpt_err) {
2603             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2604             error_free(resize_hpt_err);
2605             resize_hpt_err = NULL;
2606         } else {
2607             spapr->resize_hpt = smc->resize_hpt_default;
2608         }
2609     }
2610 
2611     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2612 
2613     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2614         /*
2615          * User requested HPT resize, but this host can't supply it.  Bail out
2616          */
2617         error_report_err(resize_hpt_err);
2618         exit(1);
2619     }
2620 
2621     spapr->rma_size = node0_size;
2622 
2623     /* With KVM, we don't actually know whether KVM supports an
2624      * unbounded RMA (PR KVM) or is limited by the hash table size
2625      * (HV KVM using VRMA), so we always assume the latter
2626      *
2627      * In that case, we also limit the initial allocations for RTAS
2628      * etc... to 256M since we have no way to know what the VRMA size
2629      * is going to be as it depends on the size of the hash table
2630      * which isn't determined yet.
2631      */
2632     if (kvm_enabled()) {
2633         spapr->vrma_adjust = 1;
2634         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2635     }
2636 
2637     /* Actually we don't support unbounded RMA anymore since we added
2638      * proper emulation of HV mode. The max we can get is 16G which
2639      * also happens to be what we configure for PAPR mode so make sure
2640      * we don't do anything bigger than that
2641      */
2642     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2643 
2644     if (spapr->rma_size > node0_size) {
2645         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2646                      spapr->rma_size);
2647         exit(1);
2648     }
2649 
2650     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2651     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2652 
2653     /*
2654      * VSMT must be set in order to be able to compute VCPU ids, ie to
2655      * call spapr_max_server_number() or spapr_vcpu_id().
2656      */
2657     spapr_set_vsmt_mode(spapr, &error_fatal);
2658 
2659     /* Set up Interrupt Controller before we create the VCPUs */
2660     spapr_irq_init(spapr, &error_fatal);
2661 
2662     /* Set up containers for ibm,client-architecture-support negotiated options
2663      */
2664     spapr->ov5 = spapr_ovec_new();
2665     spapr->ov5_cas = spapr_ovec_new();
2666 
2667     if (smc->dr_lmb_enabled) {
2668         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2669         spapr_validate_node_memory(machine, &error_fatal);
2670     }
2671 
2672     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2673 
2674     /* advertise support for dedicated HP event source to guests */
2675     if (spapr->use_hotplug_event_source) {
2676         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2677     }
2678 
2679     /* advertise support for HPT resizing */
2680     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2681         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2682     }
2683 
2684     /* advertise support for ibm,dyamic-memory-v2 */
2685     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2686 
2687     /* advertise XIVE on POWER9 machines */
2688     if (spapr->irq->ov5 & (SPAPR_OV5_XIVE_EXPLOIT | SPAPR_OV5_XIVE_BOTH)) {
2689         if (ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00,
2690                                   0, spapr->max_compat_pvr)) {
2691             spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2692         } else if (spapr->irq->ov5 & SPAPR_OV5_XIVE_EXPLOIT) {
2693             error_report("XIVE-only machines require a POWER9 CPU");
2694             exit(1);
2695         }
2696     }
2697 
2698     /* init CPUs */
2699     spapr_init_cpus(spapr);
2700 
2701     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2702         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2703                               spapr->max_compat_pvr)) {
2704         /* KVM and TCG always allow GTSE with radix... */
2705         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2706     }
2707     /* ... but not with hash (currently). */
2708 
2709     if (kvm_enabled()) {
2710         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2711         kvmppc_enable_logical_ci_hcalls();
2712         kvmppc_enable_set_mode_hcall();
2713 
2714         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2715         kvmppc_enable_clear_ref_mod_hcalls();
2716     }
2717 
2718     /* allocate RAM */
2719     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2720                                          machine->ram_size);
2721     memory_region_add_subregion(sysmem, 0, ram);
2722 
2723     /* always allocate the device memory information */
2724     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2725 
2726     /* initialize hotplug memory address space */
2727     if (machine->ram_size < machine->maxram_size) {
2728         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2729         /*
2730          * Limit the number of hotpluggable memory slots to half the number
2731          * slots that KVM supports, leaving the other half for PCI and other
2732          * devices. However ensure that number of slots doesn't drop below 32.
2733          */
2734         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2735                            SPAPR_MAX_RAM_SLOTS;
2736 
2737         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2738             max_memslots = SPAPR_MAX_RAM_SLOTS;
2739         }
2740         if (machine->ram_slots > max_memslots) {
2741             error_report("Specified number of memory slots %"
2742                          PRIu64" exceeds max supported %d",
2743                          machine->ram_slots, max_memslots);
2744             exit(1);
2745         }
2746 
2747         machine->device_memory->base = ROUND_UP(machine->ram_size,
2748                                                 SPAPR_DEVICE_MEM_ALIGN);
2749         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2750                            "device-memory", device_mem_size);
2751         memory_region_add_subregion(sysmem, machine->device_memory->base,
2752                                     &machine->device_memory->mr);
2753     }
2754 
2755     if (smc->dr_lmb_enabled) {
2756         spapr_create_lmb_dr_connectors(spapr);
2757     }
2758 
2759     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2760     if (!filename) {
2761         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2762         exit(1);
2763     }
2764     spapr->rtas_size = get_image_size(filename);
2765     if (spapr->rtas_size < 0) {
2766         error_report("Could not get size of LPAR rtas '%s'", filename);
2767         exit(1);
2768     }
2769     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2770     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2771         error_report("Could not load LPAR rtas '%s'", filename);
2772         exit(1);
2773     }
2774     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2775         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2776                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2777         exit(1);
2778     }
2779     g_free(filename);
2780 
2781     /* Set up RTAS event infrastructure */
2782     spapr_events_init(spapr);
2783 
2784     /* Set up the RTC RTAS interfaces */
2785     spapr_rtc_create(spapr);
2786 
2787     /* Set up VIO bus */
2788     spapr->vio_bus = spapr_vio_bus_init();
2789 
2790     for (i = 0; i < serial_max_hds(); i++) {
2791         if (serial_hd(i)) {
2792             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2793         }
2794     }
2795 
2796     /* We always have at least the nvram device on VIO */
2797     spapr_create_nvram(spapr);
2798 
2799     /* Set up PCI */
2800     spapr_pci_rtas_init();
2801 
2802     phb = spapr_create_default_phb();
2803 
2804     for (i = 0; i < nb_nics; i++) {
2805         NICInfo *nd = &nd_table[i];
2806 
2807         if (!nd->model) {
2808             nd->model = g_strdup("spapr-vlan");
2809         }
2810 
2811         if (g_str_equal(nd->model, "spapr-vlan") ||
2812             g_str_equal(nd->model, "ibmveth")) {
2813             spapr_vlan_create(spapr->vio_bus, nd);
2814         } else {
2815             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2816         }
2817     }
2818 
2819     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2820         spapr_vscsi_create(spapr->vio_bus);
2821     }
2822 
2823     /* Graphics */
2824     if (spapr_vga_init(phb->bus, &error_fatal)) {
2825         spapr->has_graphics = true;
2826         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2827     }
2828 
2829     if (machine->usb) {
2830         if (smc->use_ohci_by_default) {
2831             pci_create_simple(phb->bus, -1, "pci-ohci");
2832         } else {
2833             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2834         }
2835 
2836         if (spapr->has_graphics) {
2837             USBBus *usb_bus = usb_bus_find(-1);
2838 
2839             usb_create_simple(usb_bus, "usb-kbd");
2840             usb_create_simple(usb_bus, "usb-mouse");
2841         }
2842     }
2843 
2844     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2845         error_report(
2846             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2847             MIN_RMA_SLOF);
2848         exit(1);
2849     }
2850 
2851     if (kernel_filename) {
2852         uint64_t lowaddr = 0;
2853 
2854         spapr->kernel_size = load_elf(kernel_filename, NULL,
2855                                       translate_kernel_address, NULL,
2856                                       NULL, &lowaddr, NULL, 1,
2857                                       PPC_ELF_MACHINE, 0, 0);
2858         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2859             spapr->kernel_size = load_elf(kernel_filename, NULL,
2860                                           translate_kernel_address, NULL, NULL,
2861                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2862                                           0, 0);
2863             spapr->kernel_le = spapr->kernel_size > 0;
2864         }
2865         if (spapr->kernel_size < 0) {
2866             error_report("error loading %s: %s", kernel_filename,
2867                          load_elf_strerror(spapr->kernel_size));
2868             exit(1);
2869         }
2870 
2871         /* load initrd */
2872         if (initrd_filename) {
2873             /* Try to locate the initrd in the gap between the kernel
2874              * and the firmware. Add a bit of space just in case
2875              */
2876             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2877                                   + 0x1ffff) & ~0xffff;
2878             spapr->initrd_size = load_image_targphys(initrd_filename,
2879                                                      spapr->initrd_base,
2880                                                      load_limit
2881                                                      - spapr->initrd_base);
2882             if (spapr->initrd_size < 0) {
2883                 error_report("could not load initial ram disk '%s'",
2884                              initrd_filename);
2885                 exit(1);
2886             }
2887         }
2888     }
2889 
2890     if (bios_name == NULL) {
2891         bios_name = FW_FILE_NAME;
2892     }
2893     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2894     if (!filename) {
2895         error_report("Could not find LPAR firmware '%s'", bios_name);
2896         exit(1);
2897     }
2898     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2899     if (fw_size <= 0) {
2900         error_report("Could not load LPAR firmware '%s'", filename);
2901         exit(1);
2902     }
2903     g_free(filename);
2904 
2905     /* FIXME: Should register things through the MachineState's qdev
2906      * interface, this is a legacy from the sPAPREnvironment structure
2907      * which predated MachineState but had a similar function */
2908     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2909     register_savevm_live(NULL, "spapr/htab", -1, 1,
2910                          &savevm_htab_handlers, spapr);
2911 
2912     qemu_register_boot_set(spapr_boot_set, spapr);
2913 
2914     if (kvm_enabled()) {
2915         /* to stop and start vmclock */
2916         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2917                                          &spapr->tb);
2918 
2919         kvmppc_spapr_enable_inkernel_multitce();
2920     }
2921 }
2922 
2923 static int spapr_kvm_type(const char *vm_type)
2924 {
2925     if (!vm_type) {
2926         return 0;
2927     }
2928 
2929     if (!strcmp(vm_type, "HV")) {
2930         return 1;
2931     }
2932 
2933     if (!strcmp(vm_type, "PR")) {
2934         return 2;
2935     }
2936 
2937     error_report("Unknown kvm-type specified '%s'", vm_type);
2938     exit(1);
2939 }
2940 
2941 /*
2942  * Implementation of an interface to adjust firmware path
2943  * for the bootindex property handling.
2944  */
2945 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2946                                    DeviceState *dev)
2947 {
2948 #define CAST(type, obj, name) \
2949     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2950     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2951     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2952     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2953 
2954     if (d) {
2955         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2956         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2957         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2958 
2959         if (spapr) {
2960             /*
2961              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2962              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
2963              * 0x8000 | (target << 8) | (bus << 5) | lun
2964              * (see the "Logical unit addressing format" table in SAM5)
2965              */
2966             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
2967             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2968                                    (uint64_t)id << 48);
2969         } else if (virtio) {
2970             /*
2971              * We use SRP luns of the form 01000000 | (target << 8) | lun
2972              * in the top 32 bits of the 64-bit LUN
2973              * Note: the quote above is from SLOF and it is wrong,
2974              * the actual binding is:
2975              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2976              */
2977             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2978             if (d->lun >= 256) {
2979                 /* Use the LUN "flat space addressing method" */
2980                 id |= 0x4000;
2981             }
2982             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2983                                    (uint64_t)id << 32);
2984         } else if (usb) {
2985             /*
2986              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2987              * in the top 32 bits of the 64-bit LUN
2988              */
2989             unsigned usb_port = atoi(usb->port->path);
2990             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2991             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2992                                    (uint64_t)id << 32);
2993         }
2994     }
2995 
2996     /*
2997      * SLOF probes the USB devices, and if it recognizes that the device is a
2998      * storage device, it changes its name to "storage" instead of "usb-host",
2999      * and additionally adds a child node for the SCSI LUN, so the correct
3000      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3001      */
3002     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3003         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3004         if (usb_host_dev_is_scsi_storage(usbdev)) {
3005             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3006         }
3007     }
3008 
3009     if (phb) {
3010         /* Replace "pci" with "pci@800000020000000" */
3011         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3012     }
3013 
3014     if (vsc) {
3015         /* Same logic as virtio above */
3016         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3017         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3018     }
3019 
3020     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3021         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3022         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3023         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3024     }
3025 
3026     return NULL;
3027 }
3028 
3029 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3030 {
3031     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3032 
3033     return g_strdup(spapr->kvm_type);
3034 }
3035 
3036 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3037 {
3038     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3039 
3040     g_free(spapr->kvm_type);
3041     spapr->kvm_type = g_strdup(value);
3042 }
3043 
3044 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3045 {
3046     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3047 
3048     return spapr->use_hotplug_event_source;
3049 }
3050 
3051 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3052                                             Error **errp)
3053 {
3054     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3055 
3056     spapr->use_hotplug_event_source = value;
3057 }
3058 
3059 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3060 {
3061     return true;
3062 }
3063 
3064 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3065 {
3066     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3067 
3068     switch (spapr->resize_hpt) {
3069     case SPAPR_RESIZE_HPT_DEFAULT:
3070         return g_strdup("default");
3071     case SPAPR_RESIZE_HPT_DISABLED:
3072         return g_strdup("disabled");
3073     case SPAPR_RESIZE_HPT_ENABLED:
3074         return g_strdup("enabled");
3075     case SPAPR_RESIZE_HPT_REQUIRED:
3076         return g_strdup("required");
3077     }
3078     g_assert_not_reached();
3079 }
3080 
3081 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3082 {
3083     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3084 
3085     if (strcmp(value, "default") == 0) {
3086         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3087     } else if (strcmp(value, "disabled") == 0) {
3088         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3089     } else if (strcmp(value, "enabled") == 0) {
3090         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3091     } else if (strcmp(value, "required") == 0) {
3092         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3093     } else {
3094         error_setg(errp, "Bad value for \"resize-hpt\" property");
3095     }
3096 }
3097 
3098 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3099                                    void *opaque, Error **errp)
3100 {
3101     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3102 }
3103 
3104 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3105                                    void *opaque, Error **errp)
3106 {
3107     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3108 }
3109 
3110 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3111 {
3112     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3113 
3114     if (spapr->irq == &spapr_irq_xics_legacy) {
3115         return g_strdup("legacy");
3116     } else if (spapr->irq == &spapr_irq_xics) {
3117         return g_strdup("xics");
3118     } else if (spapr->irq == &spapr_irq_xive) {
3119         return g_strdup("xive");
3120     } else if (spapr->irq == &spapr_irq_dual) {
3121         return g_strdup("dual");
3122     }
3123     g_assert_not_reached();
3124 }
3125 
3126 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3127 {
3128     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3129 
3130     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3131         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3132         return;
3133     }
3134 
3135     /* The legacy IRQ backend can not be set */
3136     if (strcmp(value, "xics") == 0) {
3137         spapr->irq = &spapr_irq_xics;
3138     } else if (strcmp(value, "xive") == 0) {
3139         spapr->irq = &spapr_irq_xive;
3140     } else if (strcmp(value, "dual") == 0) {
3141         spapr->irq = &spapr_irq_dual;
3142     } else {
3143         error_setg(errp, "Bad value for \"ic-mode\" property");
3144     }
3145 }
3146 
3147 static void spapr_instance_init(Object *obj)
3148 {
3149     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3150     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3151 
3152     spapr->htab_fd = -1;
3153     spapr->use_hotplug_event_source = true;
3154     object_property_add_str(obj, "kvm-type",
3155                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3156     object_property_set_description(obj, "kvm-type",
3157                                     "Specifies the KVM virtualization mode (HV, PR)",
3158                                     NULL);
3159     object_property_add_bool(obj, "modern-hotplug-events",
3160                             spapr_get_modern_hotplug_events,
3161                             spapr_set_modern_hotplug_events,
3162                             NULL);
3163     object_property_set_description(obj, "modern-hotplug-events",
3164                                     "Use dedicated hotplug event mechanism in"
3165                                     " place of standard EPOW events when possible"
3166                                     " (required for memory hot-unplug support)",
3167                                     NULL);
3168     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3169                             "Maximum permitted CPU compatibility mode",
3170                             &error_fatal);
3171 
3172     object_property_add_str(obj, "resize-hpt",
3173                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3174     object_property_set_description(obj, "resize-hpt",
3175                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3176                                     NULL);
3177     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3178                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3179     object_property_set_description(obj, "vsmt",
3180                                     "Virtual SMT: KVM behaves as if this were"
3181                                     " the host's SMT mode", &error_abort);
3182     object_property_add_bool(obj, "vfio-no-msix-emulation",
3183                              spapr_get_msix_emulation, NULL, NULL);
3184 
3185     /* The machine class defines the default interrupt controller mode */
3186     spapr->irq = smc->irq;
3187     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3188                             spapr_set_ic_mode, NULL);
3189     object_property_set_description(obj, "ic-mode",
3190                  "Specifies the interrupt controller mode (xics, xive, dual)",
3191                  NULL);
3192 }
3193 
3194 static void spapr_machine_finalizefn(Object *obj)
3195 {
3196     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3197 
3198     g_free(spapr->kvm_type);
3199 }
3200 
3201 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3202 {
3203     cpu_synchronize_state(cs);
3204     ppc_cpu_do_system_reset(cs);
3205 }
3206 
3207 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3208 {
3209     CPUState *cs;
3210 
3211     CPU_FOREACH(cs) {
3212         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3213     }
3214 }
3215 
3216 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3217                            uint32_t node, bool dedicated_hp_event_source,
3218                            Error **errp)
3219 {
3220     sPAPRDRConnector *drc;
3221     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3222     int i, fdt_offset, fdt_size;
3223     void *fdt;
3224     uint64_t addr = addr_start;
3225     bool hotplugged = spapr_drc_hotplugged(dev);
3226     Error *local_err = NULL;
3227 
3228     for (i = 0; i < nr_lmbs; i++) {
3229         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3230                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3231         g_assert(drc);
3232 
3233         fdt = create_device_tree(&fdt_size);
3234         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
3235                                                 SPAPR_MEMORY_BLOCK_SIZE);
3236 
3237         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3238         if (local_err) {
3239             while (addr > addr_start) {
3240                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3241                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3242                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3243                 spapr_drc_detach(drc);
3244             }
3245             g_free(fdt);
3246             error_propagate(errp, local_err);
3247             return;
3248         }
3249         if (!hotplugged) {
3250             spapr_drc_reset(drc);
3251         }
3252         addr += SPAPR_MEMORY_BLOCK_SIZE;
3253     }
3254     /* send hotplug notification to the
3255      * guest only in case of hotplugged memory
3256      */
3257     if (hotplugged) {
3258         if (dedicated_hp_event_source) {
3259             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3260                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3261             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3262                                                    nr_lmbs,
3263                                                    spapr_drc_index(drc));
3264         } else {
3265             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3266                                            nr_lmbs);
3267         }
3268     }
3269 }
3270 
3271 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3272                               Error **errp)
3273 {
3274     Error *local_err = NULL;
3275     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3276     PCDIMMDevice *dimm = PC_DIMM(dev);
3277     uint64_t size, addr;
3278     uint32_t node;
3279 
3280     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3281 
3282     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3283     if (local_err) {
3284         goto out;
3285     }
3286 
3287     addr = object_property_get_uint(OBJECT(dimm),
3288                                     PC_DIMM_ADDR_PROP, &local_err);
3289     if (local_err) {
3290         goto out_unplug;
3291     }
3292 
3293     node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP,
3294                                     &error_abort);
3295     spapr_add_lmbs(dev, addr, size, node,
3296                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3297                    &local_err);
3298     if (local_err) {
3299         goto out_unplug;
3300     }
3301 
3302     return;
3303 
3304 out_unplug:
3305     pc_dimm_unplug(dimm, MACHINE(ms));
3306 out:
3307     error_propagate(errp, local_err);
3308 }
3309 
3310 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3311                                   Error **errp)
3312 {
3313     const sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3314     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3315     PCDIMMDevice *dimm = PC_DIMM(dev);
3316     Error *local_err = NULL;
3317     uint64_t size;
3318     Object *memdev;
3319     hwaddr pagesize;
3320 
3321     if (!smc->dr_lmb_enabled) {
3322         error_setg(errp, "Memory hotplug not supported for this machine");
3323         return;
3324     }
3325 
3326     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3327     if (local_err) {
3328         error_propagate(errp, local_err);
3329         return;
3330     }
3331 
3332     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3333         error_setg(errp, "Hotplugged memory size must be a multiple of "
3334                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3335         return;
3336     }
3337 
3338     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3339                                       &error_abort);
3340     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3341     spapr_check_pagesize(spapr, pagesize, &local_err);
3342     if (local_err) {
3343         error_propagate(errp, local_err);
3344         return;
3345     }
3346 
3347     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3348 }
3349 
3350 struct sPAPRDIMMState {
3351     PCDIMMDevice *dimm;
3352     uint32_t nr_lmbs;
3353     QTAILQ_ENTRY(sPAPRDIMMState) next;
3354 };
3355 
3356 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3357                                                        PCDIMMDevice *dimm)
3358 {
3359     sPAPRDIMMState *dimm_state = NULL;
3360 
3361     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3362         if (dimm_state->dimm == dimm) {
3363             break;
3364         }
3365     }
3366     return dimm_state;
3367 }
3368 
3369 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3370                                                       uint32_t nr_lmbs,
3371                                                       PCDIMMDevice *dimm)
3372 {
3373     sPAPRDIMMState *ds = NULL;
3374 
3375     /*
3376      * If this request is for a DIMM whose removal had failed earlier
3377      * (due to guest's refusal to remove the LMBs), we would have this
3378      * dimm already in the pending_dimm_unplugs list. In that
3379      * case don't add again.
3380      */
3381     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3382     if (!ds) {
3383         ds = g_malloc0(sizeof(sPAPRDIMMState));
3384         ds->nr_lmbs = nr_lmbs;
3385         ds->dimm = dimm;
3386         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3387     }
3388     return ds;
3389 }
3390 
3391 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3392                                               sPAPRDIMMState *dimm_state)
3393 {
3394     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3395     g_free(dimm_state);
3396 }
3397 
3398 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3399                                                         PCDIMMDevice *dimm)
3400 {
3401     sPAPRDRConnector *drc;
3402     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3403                                                   &error_abort);
3404     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3405     uint32_t avail_lmbs = 0;
3406     uint64_t addr_start, addr;
3407     int i;
3408 
3409     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3410                                          &error_abort);
3411 
3412     addr = addr_start;
3413     for (i = 0; i < nr_lmbs; i++) {
3414         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3415                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3416         g_assert(drc);
3417         if (drc->dev) {
3418             avail_lmbs++;
3419         }
3420         addr += SPAPR_MEMORY_BLOCK_SIZE;
3421     }
3422 
3423     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3424 }
3425 
3426 /* Callback to be called during DRC release. */
3427 void spapr_lmb_release(DeviceState *dev)
3428 {
3429     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3430     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3431     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3432 
3433     /* This information will get lost if a migration occurs
3434      * during the unplug process. In this case recover it. */
3435     if (ds == NULL) {
3436         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3437         g_assert(ds);
3438         /* The DRC being examined by the caller at least must be counted */
3439         g_assert(ds->nr_lmbs);
3440     }
3441 
3442     if (--ds->nr_lmbs) {
3443         return;
3444     }
3445 
3446     /*
3447      * Now that all the LMBs have been removed by the guest, call the
3448      * unplug handler chain. This can never fail.
3449      */
3450     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3451 }
3452 
3453 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3454 {
3455     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3456     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3457 
3458     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3459     object_unparent(OBJECT(dev));
3460     spapr_pending_dimm_unplugs_remove(spapr, ds);
3461 }
3462 
3463 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3464                                         DeviceState *dev, Error **errp)
3465 {
3466     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3467     Error *local_err = NULL;
3468     PCDIMMDevice *dimm = PC_DIMM(dev);
3469     uint32_t nr_lmbs;
3470     uint64_t size, addr_start, addr;
3471     int i;
3472     sPAPRDRConnector *drc;
3473 
3474     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3475     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3476 
3477     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3478                                          &local_err);
3479     if (local_err) {
3480         goto out;
3481     }
3482 
3483     /*
3484      * An existing pending dimm state for this DIMM means that there is an
3485      * unplug operation in progress, waiting for the spapr_lmb_release
3486      * callback to complete the job (BQL can't cover that far). In this case,
3487      * bail out to avoid detaching DRCs that were already released.
3488      */
3489     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3490         error_setg(&local_err,
3491                    "Memory unplug already in progress for device %s",
3492                    dev->id);
3493         goto out;
3494     }
3495 
3496     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3497 
3498     addr = addr_start;
3499     for (i = 0; i < nr_lmbs; i++) {
3500         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3501                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3502         g_assert(drc);
3503 
3504         spapr_drc_detach(drc);
3505         addr += SPAPR_MEMORY_BLOCK_SIZE;
3506     }
3507 
3508     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3509                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3510     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3511                                               nr_lmbs, spapr_drc_index(drc));
3512 out:
3513     error_propagate(errp, local_err);
3514 }
3515 
3516 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3517                                            sPAPRMachineState *spapr)
3518 {
3519     PowerPCCPU *cpu = POWERPC_CPU(cs);
3520     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3521     int id = spapr_get_vcpu_id(cpu);
3522     void *fdt;
3523     int offset, fdt_size;
3524     char *nodename;
3525 
3526     fdt = create_device_tree(&fdt_size);
3527     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3528     offset = fdt_add_subnode(fdt, 0, nodename);
3529 
3530     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3531     g_free(nodename);
3532 
3533     *fdt_offset = offset;
3534     return fdt;
3535 }
3536 
3537 /* Callback to be called during DRC release. */
3538 void spapr_core_release(DeviceState *dev)
3539 {
3540     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3541 
3542     /* Call the unplug handler chain. This can never fail. */
3543     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3544 }
3545 
3546 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3547 {
3548     MachineState *ms = MACHINE(hotplug_dev);
3549     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3550     CPUCore *cc = CPU_CORE(dev);
3551     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3552 
3553     if (smc->pre_2_10_has_unused_icps) {
3554         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3555         int i;
3556 
3557         for (i = 0; i < cc->nr_threads; i++) {
3558             CPUState *cs = CPU(sc->threads[i]);
3559 
3560             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3561         }
3562     }
3563 
3564     assert(core_slot);
3565     core_slot->cpu = NULL;
3566     object_unparent(OBJECT(dev));
3567 }
3568 
3569 static
3570 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3571                                Error **errp)
3572 {
3573     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3574     int index;
3575     sPAPRDRConnector *drc;
3576     CPUCore *cc = CPU_CORE(dev);
3577 
3578     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3579         error_setg(errp, "Unable to find CPU core with core-id: %d",
3580                    cc->core_id);
3581         return;
3582     }
3583     if (index == 0) {
3584         error_setg(errp, "Boot CPU core may not be unplugged");
3585         return;
3586     }
3587 
3588     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3589                           spapr_vcpu_id(spapr, cc->core_id));
3590     g_assert(drc);
3591 
3592     spapr_drc_detach(drc);
3593 
3594     spapr_hotplug_req_remove_by_index(drc);
3595 }
3596 
3597 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3598                             Error **errp)
3599 {
3600     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3601     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3602     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3603     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3604     CPUCore *cc = CPU_CORE(dev);
3605     CPUState *cs = CPU(core->threads[0]);
3606     sPAPRDRConnector *drc;
3607     Error *local_err = NULL;
3608     CPUArchId *core_slot;
3609     int index;
3610     bool hotplugged = spapr_drc_hotplugged(dev);
3611 
3612     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3613     if (!core_slot) {
3614         error_setg(errp, "Unable to find CPU core with core-id: %d",
3615                    cc->core_id);
3616         return;
3617     }
3618     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3619                           spapr_vcpu_id(spapr, cc->core_id));
3620 
3621     g_assert(drc || !mc->has_hotpluggable_cpus);
3622 
3623     if (drc) {
3624         void *fdt;
3625         int fdt_offset;
3626 
3627         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3628 
3629         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3630         if (local_err) {
3631             g_free(fdt);
3632             error_propagate(errp, local_err);
3633             return;
3634         }
3635 
3636         if (hotplugged) {
3637             /*
3638              * Send hotplug notification interrupt to the guest only
3639              * in case of hotplugged CPUs.
3640              */
3641             spapr_hotplug_req_add_by_index(drc);
3642         } else {
3643             spapr_drc_reset(drc);
3644         }
3645     }
3646 
3647     core_slot->cpu = OBJECT(dev);
3648 
3649     if (smc->pre_2_10_has_unused_icps) {
3650         int i;
3651 
3652         for (i = 0; i < cc->nr_threads; i++) {
3653             cs = CPU(core->threads[i]);
3654             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3655         }
3656     }
3657 }
3658 
3659 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3660                                 Error **errp)
3661 {
3662     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3663     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3664     Error *local_err = NULL;
3665     CPUCore *cc = CPU_CORE(dev);
3666     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3667     const char *type = object_get_typename(OBJECT(dev));
3668     CPUArchId *core_slot;
3669     int index;
3670 
3671     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3672         error_setg(&local_err, "CPU hotplug not supported for this machine");
3673         goto out;
3674     }
3675 
3676     if (strcmp(base_core_type, type)) {
3677         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3678         goto out;
3679     }
3680 
3681     if (cc->core_id % smp_threads) {
3682         error_setg(&local_err, "invalid core id %d", cc->core_id);
3683         goto out;
3684     }
3685 
3686     /*
3687      * In general we should have homogeneous threads-per-core, but old
3688      * (pre hotplug support) machine types allow the last core to have
3689      * reduced threads as a compatibility hack for when we allowed
3690      * total vcpus not a multiple of threads-per-core.
3691      */
3692     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3693         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3694                    cc->nr_threads, smp_threads);
3695         goto out;
3696     }
3697 
3698     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3699     if (!core_slot) {
3700         error_setg(&local_err, "core id %d out of range", cc->core_id);
3701         goto out;
3702     }
3703 
3704     if (core_slot->cpu) {
3705         error_setg(&local_err, "core %d already populated", cc->core_id);
3706         goto out;
3707     }
3708 
3709     numa_cpu_pre_plug(core_slot, dev, &local_err);
3710 
3711 out:
3712     error_propagate(errp, local_err);
3713 }
3714 
3715 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3716                                       DeviceState *dev, Error **errp)
3717 {
3718     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3719         spapr_memory_plug(hotplug_dev, dev, errp);
3720     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3721         spapr_core_plug(hotplug_dev, dev, errp);
3722     }
3723 }
3724 
3725 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3726                                         DeviceState *dev, Error **errp)
3727 {
3728     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3729         spapr_memory_unplug(hotplug_dev, dev);
3730     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3731         spapr_core_unplug(hotplug_dev, dev);
3732     }
3733 }
3734 
3735 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3736                                                 DeviceState *dev, Error **errp)
3737 {
3738     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3739     MachineClass *mc = MACHINE_GET_CLASS(sms);
3740 
3741     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3742         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3743             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3744         } else {
3745             /* NOTE: this means there is a window after guest reset, prior to
3746              * CAS negotiation, where unplug requests will fail due to the
3747              * capability not being detected yet. This is a bit different than
3748              * the case with PCI unplug, where the events will be queued and
3749              * eventually handled by the guest after boot
3750              */
3751             error_setg(errp, "Memory hot unplug not supported for this guest");
3752         }
3753     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3754         if (!mc->has_hotpluggable_cpus) {
3755             error_setg(errp, "CPU hot unplug not supported on this machine");
3756             return;
3757         }
3758         spapr_core_unplug_request(hotplug_dev, dev, errp);
3759     }
3760 }
3761 
3762 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3763                                           DeviceState *dev, Error **errp)
3764 {
3765     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3766         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3767     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3768         spapr_core_pre_plug(hotplug_dev, dev, errp);
3769     }
3770 }
3771 
3772 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3773                                                  DeviceState *dev)
3774 {
3775     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3776         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3777         return HOTPLUG_HANDLER(machine);
3778     }
3779     return NULL;
3780 }
3781 
3782 static CpuInstanceProperties
3783 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3784 {
3785     CPUArchId *core_slot;
3786     MachineClass *mc = MACHINE_GET_CLASS(machine);
3787 
3788     /* make sure possible_cpu are intialized */
3789     mc->possible_cpu_arch_ids(machine);
3790     /* get CPU core slot containing thread that matches cpu_index */
3791     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3792     assert(core_slot);
3793     return core_slot->props;
3794 }
3795 
3796 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3797 {
3798     return idx / smp_cores % nb_numa_nodes;
3799 }
3800 
3801 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3802 {
3803     int i;
3804     const char *core_type;
3805     int spapr_max_cores = max_cpus / smp_threads;
3806     MachineClass *mc = MACHINE_GET_CLASS(machine);
3807 
3808     if (!mc->has_hotpluggable_cpus) {
3809         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3810     }
3811     if (machine->possible_cpus) {
3812         assert(machine->possible_cpus->len == spapr_max_cores);
3813         return machine->possible_cpus;
3814     }
3815 
3816     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3817     if (!core_type) {
3818         error_report("Unable to find sPAPR CPU Core definition");
3819         exit(1);
3820     }
3821 
3822     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3823                              sizeof(CPUArchId) * spapr_max_cores);
3824     machine->possible_cpus->len = spapr_max_cores;
3825     for (i = 0; i < machine->possible_cpus->len; i++) {
3826         int core_id = i * smp_threads;
3827 
3828         machine->possible_cpus->cpus[i].type = core_type;
3829         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3830         machine->possible_cpus->cpus[i].arch_id = core_id;
3831         machine->possible_cpus->cpus[i].props.has_core_id = true;
3832         machine->possible_cpus->cpus[i].props.core_id = core_id;
3833     }
3834     return machine->possible_cpus;
3835 }
3836 
3837 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3838                                 uint64_t *buid, hwaddr *pio,
3839                                 hwaddr *mmio32, hwaddr *mmio64,
3840                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3841 {
3842     /*
3843      * New-style PHB window placement.
3844      *
3845      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3846      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3847      * windows.
3848      *
3849      * Some guest kernels can't work with MMIO windows above 1<<46
3850      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3851      *
3852      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3853      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3854      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3855      * 1TiB 64-bit MMIO windows for each PHB.
3856      */
3857     const uint64_t base_buid = 0x800000020000000ULL;
3858     int i;
3859 
3860     /* Sanity check natural alignments */
3861     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3862     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3863     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3864     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3865     /* Sanity check bounds */
3866     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3867                       SPAPR_PCI_MEM32_WIN_SIZE);
3868     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3869                       SPAPR_PCI_MEM64_WIN_SIZE);
3870 
3871     if (index >= SPAPR_MAX_PHBS) {
3872         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3873                    SPAPR_MAX_PHBS - 1);
3874         return;
3875     }
3876 
3877     *buid = base_buid + index;
3878     for (i = 0; i < n_dma; ++i) {
3879         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3880     }
3881 
3882     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3883     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3884     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3885 }
3886 
3887 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3888 {
3889     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3890 
3891     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3892 }
3893 
3894 static void spapr_ics_resend(XICSFabric *dev)
3895 {
3896     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3897 
3898     ics_resend(spapr->ics);
3899 }
3900 
3901 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3902 {
3903     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3904 
3905     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
3906 }
3907 
3908 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3909                                  Monitor *mon)
3910 {
3911     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3912 
3913     spapr->irq->print_info(spapr, mon);
3914 }
3915 
3916 int spapr_get_vcpu_id(PowerPCCPU *cpu)
3917 {
3918     return cpu->vcpu_id;
3919 }
3920 
3921 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
3922 {
3923     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3924     int vcpu_id;
3925 
3926     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
3927 
3928     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
3929         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
3930         error_append_hint(errp, "Adjust the number of cpus to %d "
3931                           "or try to raise the number of threads per core\n",
3932                           vcpu_id * smp_threads / spapr->vsmt);
3933         return;
3934     }
3935 
3936     cpu->vcpu_id = vcpu_id;
3937 }
3938 
3939 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3940 {
3941     CPUState *cs;
3942 
3943     CPU_FOREACH(cs) {
3944         PowerPCCPU *cpu = POWERPC_CPU(cs);
3945 
3946         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
3947             return cpu;
3948         }
3949     }
3950 
3951     return NULL;
3952 }
3953 
3954 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3955 {
3956     MachineClass *mc = MACHINE_CLASS(oc);
3957     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3958     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3959     NMIClass *nc = NMI_CLASS(oc);
3960     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3961     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3962     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3963     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3964 
3965     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3966     mc->ignore_boot_device_suffixes = true;
3967 
3968     /*
3969      * We set up the default / latest behaviour here.  The class_init
3970      * functions for the specific versioned machine types can override
3971      * these details for backwards compatibility
3972      */
3973     mc->init = spapr_machine_init;
3974     mc->reset = spapr_machine_reset;
3975     mc->block_default_type = IF_SCSI;
3976     mc->max_cpus = 1024;
3977     mc->no_parallel = 1;
3978     mc->default_boot_order = "";
3979     mc->default_ram_size = 512 * MiB;
3980     mc->default_display = "std";
3981     mc->kvm_type = spapr_kvm_type;
3982     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
3983     mc->pci_allow_0_address = true;
3984     assert(!mc->get_hotplug_handler);
3985     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3986     hc->pre_plug = spapr_machine_device_pre_plug;
3987     hc->plug = spapr_machine_device_plug;
3988     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3989     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3990     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3991     hc->unplug_request = spapr_machine_device_unplug_request;
3992     hc->unplug = spapr_machine_device_unplug;
3993 
3994     smc->dr_lmb_enabled = true;
3995     smc->update_dt_enabled = true;
3996     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
3997     mc->has_hotpluggable_cpus = true;
3998     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3999     fwc->get_dev_path = spapr_get_fw_dev_path;
4000     nc->nmi_monitor_handler = spapr_nmi;
4001     smc->phb_placement = spapr_phb_placement;
4002     vhc->hypercall = emulate_spapr_hypercall;
4003     vhc->hpt_mask = spapr_hpt_mask;
4004     vhc->map_hptes = spapr_map_hptes;
4005     vhc->unmap_hptes = spapr_unmap_hptes;
4006     vhc->store_hpte = spapr_store_hpte;
4007     vhc->get_patbe = spapr_get_patbe;
4008     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4009     xic->ics_get = spapr_ics_get;
4010     xic->ics_resend = spapr_ics_resend;
4011     xic->icp_get = spapr_icp_get;
4012     ispc->print_info = spapr_pic_print_info;
4013     /* Force NUMA node memory size to be a multiple of
4014      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4015      * in which LMBs are represented and hot-added
4016      */
4017     mc->numa_mem_align_shift = 28;
4018 
4019     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4020     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4021     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4022     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4023     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4024     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4025     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4026     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4027     spapr_caps_add_properties(smc, &error_abort);
4028     smc->irq = &spapr_irq_xics;
4029 }
4030 
4031 static const TypeInfo spapr_machine_info = {
4032     .name          = TYPE_SPAPR_MACHINE,
4033     .parent        = TYPE_MACHINE,
4034     .abstract      = true,
4035     .instance_size = sizeof(sPAPRMachineState),
4036     .instance_init = spapr_instance_init,
4037     .instance_finalize = spapr_machine_finalizefn,
4038     .class_size    = sizeof(sPAPRMachineClass),
4039     .class_init    = spapr_machine_class_init,
4040     .interfaces = (InterfaceInfo[]) {
4041         { TYPE_FW_PATH_PROVIDER },
4042         { TYPE_NMI },
4043         { TYPE_HOTPLUG_HANDLER },
4044         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4045         { TYPE_XICS_FABRIC },
4046         { TYPE_INTERRUPT_STATS_PROVIDER },
4047         { }
4048     },
4049 };
4050 
4051 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4052     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4053                                                     void *data)      \
4054     {                                                                \
4055         MachineClass *mc = MACHINE_CLASS(oc);                        \
4056         spapr_machine_##suffix##_class_options(mc);                  \
4057         if (latest) {                                                \
4058             mc->alias = "pseries";                                   \
4059             mc->is_default = 1;                                      \
4060         }                                                            \
4061     }                                                                \
4062     static const TypeInfo spapr_machine_##suffix##_info = {          \
4063         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4064         .parent = TYPE_SPAPR_MACHINE,                                \
4065         .class_init = spapr_machine_##suffix##_class_init,           \
4066     };                                                               \
4067     static void spapr_machine_register_##suffix(void)                \
4068     {                                                                \
4069         type_register(&spapr_machine_##suffix##_info);               \
4070     }                                                                \
4071     type_init(spapr_machine_register_##suffix)
4072 
4073 /*
4074  * pseries-4.0
4075  */
4076 static void spapr_machine_4_0_class_options(MachineClass *mc)
4077 {
4078     /* Defaults for the latest behaviour inherited from the base class */
4079 }
4080 
4081 DEFINE_SPAPR_MACHINE(4_0, "4.0", true);
4082 
4083 /*
4084  * pseries-3.1
4085  */
4086 static void spapr_machine_3_1_class_options(MachineClass *mc)
4087 {
4088     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4089 
4090     spapr_machine_4_0_class_options(mc);
4091     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4092     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4093     smc->update_dt_enabled = false;
4094 }
4095 
4096 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4097 
4098 /*
4099  * pseries-3.0
4100  */
4101 
4102 static void spapr_machine_3_0_class_options(MachineClass *mc)
4103 {
4104     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4105 
4106     spapr_machine_3_1_class_options(mc);
4107     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4108 
4109     smc->legacy_irq_allocation = true;
4110     smc->irq = &spapr_irq_xics_legacy;
4111 }
4112 
4113 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4114 
4115 /*
4116  * pseries-2.12
4117  */
4118 static void spapr_machine_2_12_class_options(MachineClass *mc)
4119 {
4120     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4121     static GlobalProperty compat[] = {
4122         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4123         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4124     };
4125 
4126     spapr_machine_3_0_class_options(mc);
4127     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4128     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4129 
4130     /* We depend on kvm_enabled() to choose a default value for the
4131      * hpt-max-page-size capability. Of course we can't do it here
4132      * because this is too early and the HW accelerator isn't initialzed
4133      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4134      */
4135     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4136 }
4137 
4138 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4139 
4140 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4141 {
4142     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4143 
4144     spapr_machine_2_12_class_options(mc);
4145     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4146     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4147     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4148 }
4149 
4150 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4151 
4152 /*
4153  * pseries-2.11
4154  */
4155 
4156 static void spapr_machine_2_11_class_options(MachineClass *mc)
4157 {
4158     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4159 
4160     spapr_machine_2_12_class_options(mc);
4161     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4162     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4163 }
4164 
4165 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4166 
4167 /*
4168  * pseries-2.10
4169  */
4170 
4171 static void spapr_machine_2_10_class_options(MachineClass *mc)
4172 {
4173     spapr_machine_2_11_class_options(mc);
4174     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4175 }
4176 
4177 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4178 
4179 /*
4180  * pseries-2.9
4181  */
4182 
4183 static void spapr_machine_2_9_class_options(MachineClass *mc)
4184 {
4185     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4186     static GlobalProperty compat[] = {
4187         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4188     };
4189 
4190     spapr_machine_2_10_class_options(mc);
4191     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4192     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4193     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4194     smc->pre_2_10_has_unused_icps = true;
4195     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4196 }
4197 
4198 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4199 
4200 /*
4201  * pseries-2.8
4202  */
4203 
4204 static void spapr_machine_2_8_class_options(MachineClass *mc)
4205 {
4206     static GlobalProperty compat[] = {
4207         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4208     };
4209 
4210     spapr_machine_2_9_class_options(mc);
4211     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4212     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4213     mc->numa_mem_align_shift = 23;
4214 }
4215 
4216 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4217 
4218 /*
4219  * pseries-2.7
4220  */
4221 
4222 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4223                               uint64_t *buid, hwaddr *pio,
4224                               hwaddr *mmio32, hwaddr *mmio64,
4225                               unsigned n_dma, uint32_t *liobns, Error **errp)
4226 {
4227     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4228     const uint64_t base_buid = 0x800000020000000ULL;
4229     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4230     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4231     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4232     const uint32_t max_index = 255;
4233     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4234 
4235     uint64_t ram_top = MACHINE(spapr)->ram_size;
4236     hwaddr phb0_base, phb_base;
4237     int i;
4238 
4239     /* Do we have device memory? */
4240     if (MACHINE(spapr)->maxram_size > ram_top) {
4241         /* Can't just use maxram_size, because there may be an
4242          * alignment gap between normal and device memory regions
4243          */
4244         ram_top = MACHINE(spapr)->device_memory->base +
4245             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4246     }
4247 
4248     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4249 
4250     if (index > max_index) {
4251         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4252                    max_index);
4253         return;
4254     }
4255 
4256     *buid = base_buid + index;
4257     for (i = 0; i < n_dma; ++i) {
4258         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4259     }
4260 
4261     phb_base = phb0_base + index * phb_spacing;
4262     *pio = phb_base + pio_offset;
4263     *mmio32 = phb_base + mmio_offset;
4264     /*
4265      * We don't set the 64-bit MMIO window, relying on the PHB's
4266      * fallback behaviour of automatically splitting a large "32-bit"
4267      * window into contiguous 32-bit and 64-bit windows
4268      */
4269 }
4270 
4271 static void spapr_machine_2_7_class_options(MachineClass *mc)
4272 {
4273     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4274     static GlobalProperty compat[] = {
4275         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4276         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4277         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4278         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4279     };
4280 
4281     spapr_machine_2_8_class_options(mc);
4282     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4283     mc->default_machine_opts = "modern-hotplug-events=off";
4284     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4285     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4286     smc->phb_placement = phb_placement_2_7;
4287 }
4288 
4289 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4290 
4291 /*
4292  * pseries-2.6
4293  */
4294 
4295 static void spapr_machine_2_6_class_options(MachineClass *mc)
4296 {
4297     static GlobalProperty compat[] = {
4298         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4299     };
4300 
4301     spapr_machine_2_7_class_options(mc);
4302     mc->has_hotpluggable_cpus = false;
4303     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4304     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4305 }
4306 
4307 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4308 
4309 /*
4310  * pseries-2.5
4311  */
4312 
4313 static void spapr_machine_2_5_class_options(MachineClass *mc)
4314 {
4315     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4316     static GlobalProperty compat[] = {
4317         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4318     };
4319 
4320     spapr_machine_2_6_class_options(mc);
4321     smc->use_ohci_by_default = true;
4322     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4323     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4324 }
4325 
4326 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4327 
4328 /*
4329  * pseries-2.4
4330  */
4331 
4332 static void spapr_machine_2_4_class_options(MachineClass *mc)
4333 {
4334     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4335 
4336     spapr_machine_2_5_class_options(mc);
4337     smc->dr_lmb_enabled = false;
4338     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4339 }
4340 
4341 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4342 
4343 /*
4344  * pseries-2.3
4345  */
4346 
4347 static void spapr_machine_2_3_class_options(MachineClass *mc)
4348 {
4349     static GlobalProperty compat[] = {
4350         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4351     };
4352     spapr_machine_2_4_class_options(mc);
4353     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4354     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4355 }
4356 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4357 
4358 /*
4359  * pseries-2.2
4360  */
4361 
4362 static void spapr_machine_2_2_class_options(MachineClass *mc)
4363 {
4364     static GlobalProperty compat[] = {
4365         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4366     };
4367 
4368     spapr_machine_2_3_class_options(mc);
4369     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4370     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4371     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4372 }
4373 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4374 
4375 /*
4376  * pseries-2.1
4377  */
4378 
4379 static void spapr_machine_2_1_class_options(MachineClass *mc)
4380 {
4381     spapr_machine_2_2_class_options(mc);
4382     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4383 }
4384 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4385 
4386 static void spapr_machine_register_types(void)
4387 {
4388     type_register_static(&spapr_machine_info);
4389 }
4390 
4391 type_init(spapr_machine_register_types)
4392