2 * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
4 * Copyright (c) 2004-2007 Fabrice Bellard
5 * Copyright (c) 2007 Jocelyn Mayer
6 * Copyright (c) 2010 David Gibson, IBM Corporation.
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:
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
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
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
38 #include "hw/fw-path-provider.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "mmu-hash64.h"
50 #include "mmu-book3s-v3.h"
51 #include "cpu-models.h"
52 #include "hw/core/cpu.h"
54 #include "hw/boards.h"
55 #include "hw/ppc/ppc.h"
56 #include "hw/loader.h"
58 #include "hw/ppc/fdt.h"
59 #include "hw/ppc/spapr.h"
60 #include "hw/ppc/spapr_vio.h"
61 #include "hw/qdev-properties.h"
62 #include "hw/pci-host/spapr.h"
63 #include "hw/pci/msi.h"
65 #include "hw/pci/pci.h"
66 #include "hw/scsi/scsi.h"
67 #include "hw/virtio/virtio-scsi.h"
68 #include "hw/virtio/vhost-scsi-common.h"
70 #include "exec/address-spaces.h"
71 #include "exec/ram_addr.h"
73 #include "qemu/config-file.h"
74 #include "qemu/error-report.h"
77 #include "hw/intc/intc.h"
79 #include "qemu/cutils.h"
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
84 #include "monitor/monitor.h"
88 /* SLOF memory layout:
90 * SLOF raw image loaded at 0, copies its romfs right below the flat
91 * device-tree, then position SLOF itself 31M below that
93 * So we set FW_OVERHEAD to 40MB which should account for all of that
96 * We load our kernel at 4M, leaving space for SLOF initial image
98 #define FDT_MAX_SIZE 0x100000
99 #define RTAS_MAX_SIZE 0x10000
100 #define RTAS_MAX_ADDR 0x80000000 /* RTAS must stay below that */
101 #define FW_MAX_SIZE 0x400000
102 #define FW_FILE_NAME "slof.bin"
103 #define FW_OVERHEAD 0x2800000
104 #define KERNEL_LOAD_ADDR FW_MAX_SIZE
106 #define MIN_RMA_SLOF 128UL
108 #define PHANDLE_INTC 0x00001111
110 /* These two functions implement the VCPU id numbering: one to compute them
111 * all and one to identify thread 0 of a VCORE. Any change to the first one
112 * is likely to have an impact on the second one, so let's keep them close.
114 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
116 MachineState *ms = MACHINE(spapr);
117 unsigned int smp_threads = ms->smp.threads;
121 (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
123 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
127 return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
132 /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133 * and newer QEMUs don't even have them. In both cases, we don't want
134 * to send anything on the wire.
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140 .name = "icp/server",
142 .minimum_version_id = 1,
143 .needed = pre_2_10_vmstate_dummy_icp_needed,
144 .fields = (VMStateField[]) {
145 VMSTATE_UNUSED(4), /* uint32_t xirr */
146 VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147 VMSTATE_UNUSED(1), /* uint8_t mfrr */
148 VMSTATE_END_OF_LIST()
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
154 vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155 (void *)(uintptr_t) i);
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
160 vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161 (void *)(uintptr_t) i);
164 int spapr_max_server_number(SpaprMachineState *spapr)
166 MachineState *ms = MACHINE(spapr);
169 return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
172 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
176 uint32_t servers_prop[smt_threads];
177 uint32_t gservers_prop[smt_threads * 2];
178 int index = spapr_get_vcpu_id(cpu);
180 if (cpu->compat_pvr) {
181 ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
187 /* Build interrupt servers and gservers properties */
188 for (i = 0; i < smt_threads; i++) {
189 servers_prop[i] = cpu_to_be32(index + i);
190 /* Hack, direct the group queues back to cpu 0 */
191 gservers_prop[i*2] = cpu_to_be32(index + i);
192 gservers_prop[i*2 + 1] = 0;
194 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
195 servers_prop, sizeof(servers_prop));
199 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
200 gservers_prop, sizeof(gservers_prop));
205 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
207 int index = spapr_get_vcpu_id(cpu);
208 uint32_t associativity[] = {cpu_to_be32(0x5),
212 cpu_to_be32(cpu->node_id),
215 /* Advertise NUMA via ibm,associativity */
216 return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
217 sizeof(associativity));
220 /* Populate the "ibm,pa-features" property */
221 static void spapr_populate_pa_features(SpaprMachineState *spapr,
223 void *fdt, int offset)
225 uint8_t pa_features_206[] = { 6, 0,
226 0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
227 uint8_t pa_features_207[] = { 24, 0,
228 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
229 0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
230 0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
231 0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
232 uint8_t pa_features_300[] = { 66, 0,
233 /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
234 /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
235 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
237 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
239 0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
240 /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
241 0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
242 /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
243 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
244 /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
245 0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
246 /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
247 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
248 /* 42: PM, 44: PC RA, 46: SC vec'd */
249 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
250 /* 48: SIMD, 50: QP BFP, 52: String */
251 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
252 /* 54: DecFP, 56: DecI, 58: SHA */
253 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
254 /* 60: NM atomic, 62: RNG */
255 0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
257 uint8_t *pa_features = NULL;
260 if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
261 pa_features = pa_features_206;
262 pa_size = sizeof(pa_features_206);
264 if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
265 pa_features = pa_features_207;
266 pa_size = sizeof(pa_features_207);
268 if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
269 pa_features = pa_features_300;
270 pa_size = sizeof(pa_features_300);
276 if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
278 * Note: we keep CI large pages off by default because a 64K capable
279 * guest provisioned with large pages might otherwise try to map a qemu
280 * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
281 * even if that qemu runs on a 4k host.
282 * We dd this bit back here if we are confident this is not an issue
284 pa_features[3] |= 0x20;
286 if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
287 pa_features[24] |= 0x80; /* Transactional memory support */
289 if (spapr->cas_pre_isa3_guest && pa_size > 40) {
290 /* Workaround for broken kernels that attempt (guest) radix
291 * mode when they can't handle it, if they see the radix bit set
292 * in pa-features. So hide it from them. */
293 pa_features[40 + 2] &= ~0x80; /* Radix MMU */
296 _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
299 static int spapr_fixup_cpu_dt(void *fdt, SpaprMachineState *spapr)
301 MachineState *ms = MACHINE(spapr);
302 int ret = 0, offset, cpus_offset;
305 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
308 PowerPCCPU *cpu = POWERPC_CPU(cs);
309 DeviceClass *dc = DEVICE_GET_CLASS(cs);
310 int index = spapr_get_vcpu_id(cpu);
311 int compat_smt = MIN(ms->smp.threads, ppc_compat_max_vthreads(cpu));
313 if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
317 snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
319 cpus_offset = fdt_path_offset(fdt, "/cpus");
320 if (cpus_offset < 0) {
321 cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
322 if (cpus_offset < 0) {
326 offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
328 offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
334 ret = fdt_setprop(fdt, offset, "ibm,pft-size",
335 pft_size_prop, sizeof(pft_size_prop));
340 if (ms->numa_state->num_nodes > 1) {
341 ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
347 ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
352 spapr_populate_pa_features(spapr, cpu, fdt, offset);
357 static hwaddr spapr_node0_size(MachineState *machine)
359 if (machine->numa_state->num_nodes) {
361 for (i = 0; i < machine->numa_state->num_nodes; ++i) {
362 if (machine->numa_state->nodes[i].node_mem) {
363 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
368 return machine->ram_size;
371 static void add_str(GString *s, const gchar *s1)
373 g_string_append_len(s, s1, strlen(s1) + 1);
376 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
379 uint32_t associativity[] = {
380 cpu_to_be32(0x4), /* length */
381 cpu_to_be32(0x0), cpu_to_be32(0x0),
382 cpu_to_be32(0x0), cpu_to_be32(nodeid)
385 uint64_t mem_reg_property[2];
388 mem_reg_property[0] = cpu_to_be64(start);
389 mem_reg_property[1] = cpu_to_be64(size);
391 sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
392 off = fdt_add_subnode(fdt, 0, mem_name);
394 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
395 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
396 sizeof(mem_reg_property))));
397 _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
398 sizeof(associativity))));
402 static int spapr_populate_memory(SpaprMachineState *spapr, void *fdt)
404 MachineState *machine = MACHINE(spapr);
405 hwaddr mem_start, node_size;
406 int i, nb_nodes = machine->numa_state->num_nodes;
407 NodeInfo *nodes = machine->numa_state->nodes;
410 /* No NUMA nodes, assume there is just one node with whole RAM */
413 ramnode.node_mem = machine->ram_size;
417 for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
418 if (!nodes[i].node_mem) {
421 if (mem_start >= machine->ram_size) {
424 node_size = nodes[i].node_mem;
425 if (node_size > machine->ram_size - mem_start) {
426 node_size = machine->ram_size - mem_start;
430 /* spapr_machine_init() checks for rma_size <= node0_size
432 spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
433 mem_start += spapr->rma_size;
434 node_size -= spapr->rma_size;
436 for ( ; node_size; ) {
437 hwaddr sizetmp = pow2floor(node_size);
439 /* mem_start != 0 here */
440 if (ctzl(mem_start) < ctzl(sizetmp)) {
441 sizetmp = 1ULL << ctzl(mem_start);
444 spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
445 node_size -= sizetmp;
446 mem_start += sizetmp;
453 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
454 SpaprMachineState *spapr)
456 MachineState *ms = MACHINE(spapr);
457 PowerPCCPU *cpu = POWERPC_CPU(cs);
458 CPUPPCState *env = &cpu->env;
459 PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
460 int index = spapr_get_vcpu_id(cpu);
461 uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
462 0xffffffff, 0xffffffff};
463 uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
464 : SPAPR_TIMEBASE_FREQ;
465 uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
466 uint32_t page_sizes_prop[64];
467 size_t page_sizes_prop_size;
468 unsigned int smp_threads = ms->smp.threads;
469 uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
470 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
471 int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
474 uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
477 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
479 drc_index = spapr_drc_index(drc);
480 _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
483 _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
484 _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
486 _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
487 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
488 env->dcache_line_size)));
489 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
490 env->dcache_line_size)));
491 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
492 env->icache_line_size)));
493 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
494 env->icache_line_size)));
496 if (pcc->l1_dcache_size) {
497 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
498 pcc->l1_dcache_size)));
500 warn_report("Unknown L1 dcache size for cpu");
502 if (pcc->l1_icache_size) {
503 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
504 pcc->l1_icache_size)));
506 warn_report("Unknown L1 icache size for cpu");
509 _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
510 _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
511 _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
512 _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
513 _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
514 _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
516 if (env->spr_cb[SPR_PURR].oea_read) {
517 _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
519 if (env->spr_cb[SPR_SPURR].oea_read) {
520 _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
523 if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
524 _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
525 segs, sizeof(segs))));
528 /* Advertise VSX (vector extensions) if available
529 * 1 == VMX / Altivec available
532 * Only CPUs for which we create core types in spapr_cpu_core.c
533 * are possible, and all of those have VMX */
534 if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
535 _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
537 _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
540 /* Advertise DFP (Decimal Floating Point) if available
541 * 0 / no property == no DFP
542 * 1 == DFP available */
543 if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
544 _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
547 page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
548 sizeof(page_sizes_prop));
549 if (page_sizes_prop_size) {
550 _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
551 page_sizes_prop, page_sizes_prop_size)));
554 spapr_populate_pa_features(spapr, cpu, fdt, offset);
556 _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
557 cs->cpu_index / vcpus_per_socket)));
559 _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
560 pft_size_prop, sizeof(pft_size_prop))));
562 if (ms->numa_state->num_nodes > 1) {
563 _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
566 _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
568 if (pcc->radix_page_info) {
569 for (i = 0; i < pcc->radix_page_info->count; i++) {
570 radix_AP_encodings[i] =
571 cpu_to_be32(pcc->radix_page_info->entries[i]);
573 _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
575 pcc->radix_page_info->count *
576 sizeof(radix_AP_encodings[0]))));
580 * We set this property to let the guest know that it can use the large
581 * decrementer and its width in bits.
583 if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
584 _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
585 pcc->lrg_decr_bits)));
588 static void spapr_populate_cpus_dt_node(void *fdt, SpaprMachineState *spapr)
597 cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
599 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
600 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
603 * We walk the CPUs in reverse order to ensure that CPU DT nodes
604 * created by fdt_add_subnode() end up in the right order in FDT
605 * for the guest kernel the enumerate the CPUs correctly.
607 * The CPU list cannot be traversed in reverse order, so we need
613 rev = g_renew(CPUState *, rev, n_cpus + 1);
617 for (i = n_cpus - 1; i >= 0; i--) {
618 CPUState *cs = rev[i];
619 PowerPCCPU *cpu = POWERPC_CPU(cs);
620 int index = spapr_get_vcpu_id(cpu);
621 DeviceClass *dc = DEVICE_GET_CLASS(cs);
624 if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
628 nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
629 offset = fdt_add_subnode(fdt, cpus_offset, nodename);
632 spapr_populate_cpu_dt(cs, fdt, offset, spapr);
638 static int spapr_rng_populate_dt(void *fdt)
643 node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
647 ret = fdt_setprop_string(fdt, node, "device_type",
648 "ibm,platform-facilities");
649 ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
650 ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
652 node = fdt_add_subnode(fdt, node, "ibm,random-v1");
656 ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
661 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
663 MemoryDeviceInfoList *info;
665 for (info = list; info; info = info->next) {
666 MemoryDeviceInfo *value = info->value;
668 if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
669 PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
671 if (addr >= pcdimm_info->addr &&
672 addr < (pcdimm_info->addr + pcdimm_info->size)) {
673 return pcdimm_info->node;
681 struct sPAPRDrconfCellV2 {
689 typedef struct DrconfCellQueue {
690 struct sPAPRDrconfCellV2 cell;
691 QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
694 static DrconfCellQueue *
695 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
696 uint32_t drc_index, uint32_t aa_index,
699 DrconfCellQueue *elem;
701 elem = g_malloc0(sizeof(*elem));
702 elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
703 elem->cell.base_addr = cpu_to_be64(base_addr);
704 elem->cell.drc_index = cpu_to_be32(drc_index);
705 elem->cell.aa_index = cpu_to_be32(aa_index);
706 elem->cell.flags = cpu_to_be32(flags);
711 /* ibm,dynamic-memory-v2 */
712 static int spapr_populate_drmem_v2(SpaprMachineState *spapr, void *fdt,
713 int offset, MemoryDeviceInfoList *dimms)
715 MachineState *machine = MACHINE(spapr);
716 uint8_t *int_buf, *cur_index;
718 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
719 uint64_t addr, cur_addr, size;
720 uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
721 uint64_t mem_end = machine->device_memory->base +
722 memory_region_size(&machine->device_memory->mr);
723 uint32_t node, buf_len, nr_entries = 0;
725 DrconfCellQueue *elem, *next;
726 MemoryDeviceInfoList *info;
727 QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
728 = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
730 /* Entry to cover RAM and the gap area */
731 elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
732 SPAPR_LMB_FLAGS_RESERVED |
733 SPAPR_LMB_FLAGS_DRC_INVALID);
734 QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
737 cur_addr = machine->device_memory->base;
738 for (info = dimms; info; info = info->next) {
739 PCDIMMDeviceInfo *di = info->value->u.dimm.data;
745 /* Entry for hot-pluggable area */
746 if (cur_addr < addr) {
747 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
749 elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
750 cur_addr, spapr_drc_index(drc), -1, 0);
751 QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
756 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
758 elem = spapr_get_drconf_cell(size / lmb_size, addr,
759 spapr_drc_index(drc), node,
760 SPAPR_LMB_FLAGS_ASSIGNED);
761 QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
763 cur_addr = addr + size;
766 /* Entry for remaining hotpluggable area */
767 if (cur_addr < mem_end) {
768 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
770 elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
771 cur_addr, spapr_drc_index(drc), -1, 0);
772 QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
776 buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
777 int_buf = cur_index = g_malloc0(buf_len);
778 *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
779 cur_index += sizeof(nr_entries);
781 QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
782 memcpy(cur_index, &elem->cell, sizeof(elem->cell));
783 cur_index += sizeof(elem->cell);
784 QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
788 ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
796 /* ibm,dynamic-memory */
797 static int spapr_populate_drmem_v1(SpaprMachineState *spapr, void *fdt,
798 int offset, MemoryDeviceInfoList *dimms)
800 MachineState *machine = MACHINE(spapr);
802 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
803 uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
804 uint32_t nr_lmbs = (machine->device_memory->base +
805 memory_region_size(&machine->device_memory->mr)) /
807 uint32_t *int_buf, *cur_index, buf_len;
810 * Allocate enough buffer size to fit in ibm,dynamic-memory
812 buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
813 cur_index = int_buf = g_malloc0(buf_len);
814 int_buf[0] = cpu_to_be32(nr_lmbs);
816 for (i = 0; i < nr_lmbs; i++) {
817 uint64_t addr = i * lmb_size;
818 uint32_t *dynamic_memory = cur_index;
820 if (i >= device_lmb_start) {
823 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
826 dynamic_memory[0] = cpu_to_be32(addr >> 32);
827 dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
828 dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
829 dynamic_memory[3] = cpu_to_be32(0); /* reserved */
830 dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
831 if (memory_region_present(get_system_memory(), addr)) {
832 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
834 dynamic_memory[5] = cpu_to_be32(0);
838 * LMB information for RMA, boot time RAM and gap b/n RAM and
839 * device memory region -- all these are marked as reserved
840 * and as having no valid DRC.
842 dynamic_memory[0] = cpu_to_be32(addr >> 32);
843 dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
844 dynamic_memory[2] = cpu_to_be32(0);
845 dynamic_memory[3] = cpu_to_be32(0); /* reserved */
846 dynamic_memory[4] = cpu_to_be32(-1);
847 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
848 SPAPR_LMB_FLAGS_DRC_INVALID);
851 cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
853 ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
862 * Adds ibm,dynamic-reconfiguration-memory node.
863 * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
864 * of this device tree node.
866 static int spapr_populate_drconf_memory(SpaprMachineState *spapr, void *fdt)
868 MachineState *machine = MACHINE(spapr);
869 int nb_numa_nodes = machine->numa_state->num_nodes;
871 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
872 uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
873 uint32_t *int_buf, *cur_index, buf_len;
874 int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
875 MemoryDeviceInfoList *dimms = NULL;
878 * Don't create the node if there is no device memory
880 if (machine->ram_size == machine->maxram_size) {
884 offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
886 ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
887 sizeof(prop_lmb_size));
892 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
897 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
902 /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
903 dimms = qmp_memory_device_list();
904 if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
905 ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
907 ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
909 qapi_free_MemoryDeviceInfoList(dimms);
915 /* ibm,associativity-lookup-arrays */
916 buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
917 cur_index = int_buf = g_malloc0(buf_len);
918 int_buf[0] = cpu_to_be32(nr_nodes);
919 int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
921 for (i = 0; i < nr_nodes; i++) {
922 uint32_t associativity[] = {
928 memcpy(cur_index, associativity, sizeof(associativity));
931 ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
932 (cur_index - int_buf) * sizeof(uint32_t));
938 static int spapr_dt_cas_updates(SpaprMachineState *spapr, void *fdt,
939 SpaprOptionVector *ov5_updates)
941 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
944 /* Generate ibm,dynamic-reconfiguration-memory node if required */
945 if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
946 g_assert(smc->dr_lmb_enabled);
947 ret = spapr_populate_drconf_memory(spapr, fdt);
953 offset = fdt_path_offset(fdt, "/chosen");
955 offset = fdt_add_subnode(fdt, 0, "chosen");
960 ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
961 "ibm,architecture-vec-5");
967 static bool spapr_hotplugged_dev_before_cas(void)
969 Object *drc_container, *obj;
970 ObjectProperty *prop;
971 ObjectPropertyIterator iter;
973 drc_container = container_get(object_get_root(), "/dr-connector");
974 object_property_iter_init(&iter, drc_container);
975 while ((prop = object_property_iter_next(&iter))) {
976 if (!strstart(prop->type, "link<", NULL)) {
979 obj = object_property_get_link(drc_container, prop->name, NULL);
980 if (spapr_drc_needed(obj)) {
987 int spapr_h_cas_compose_response(SpaprMachineState *spapr,
988 target_ulong addr, target_ulong size,
989 SpaprOptionVector *ov5_updates)
991 void *fdt, *fdt_skel;
992 SpaprDeviceTreeUpdateHeader hdr = { .version_id = 1 };
994 if (spapr_hotplugged_dev_before_cas()) {
998 if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
999 error_report("SLOF provided an unexpected CAS buffer size "
1000 TARGET_FMT_lu " (min: %zu, max: %u)",
1001 size, sizeof(hdr), FW_MAX_SIZE);
1005 size -= sizeof(hdr);
1007 /* Create skeleton */
1008 fdt_skel = g_malloc0(size);
1009 _FDT((fdt_create(fdt_skel, size)));
1010 _FDT((fdt_finish_reservemap(fdt_skel)));
1011 _FDT((fdt_begin_node(fdt_skel, "")));
1012 _FDT((fdt_end_node(fdt_skel)));
1013 _FDT((fdt_finish(fdt_skel)));
1014 fdt = g_malloc0(size);
1015 _FDT((fdt_open_into(fdt_skel, fdt, size)));
1018 /* Fixup cpu nodes */
1019 _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
1021 if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
1025 /* Pack resulting tree */
1026 _FDT((fdt_pack(fdt)));
1028 if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
1029 trace_spapr_cas_failed(size);
1033 cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
1034 cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
1035 trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
1041 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
1043 MachineState *ms = MACHINE(spapr);
1045 GString *hypertas = g_string_sized_new(256);
1046 GString *qemu_hypertas = g_string_sized_new(256);
1047 uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1048 uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
1049 memory_region_size(&MACHINE(spapr)->device_memory->mr);
1050 uint32_t lrdr_capacity[] = {
1051 cpu_to_be32(max_device_addr >> 32),
1052 cpu_to_be32(max_device_addr & 0xffffffff),
1053 0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1054 cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
1056 uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
1057 uint32_t maxdomains[] = {
1062 cpu_to_be32(spapr->gpu_numa_id),
1065 _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1068 add_str(hypertas, "hcall-pft");
1069 add_str(hypertas, "hcall-term");
1070 add_str(hypertas, "hcall-dabr");
1071 add_str(hypertas, "hcall-interrupt");
1072 add_str(hypertas, "hcall-tce");
1073 add_str(hypertas, "hcall-vio");
1074 add_str(hypertas, "hcall-splpar");
1075 add_str(hypertas, "hcall-join");
1076 add_str(hypertas, "hcall-bulk");
1077 add_str(hypertas, "hcall-set-mode");
1078 add_str(hypertas, "hcall-sprg0");
1079 add_str(hypertas, "hcall-copy");
1080 add_str(hypertas, "hcall-debug");
1081 add_str(hypertas, "hcall-vphn");
1082 add_str(qemu_hypertas, "hcall-memop1");
1084 if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1085 add_str(hypertas, "hcall-multi-tce");
1088 if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1089 add_str(hypertas, "hcall-hpt-resize");
1092 _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1093 hypertas->str, hypertas->len));
1094 g_string_free(hypertas, TRUE);
1095 _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1096 qemu_hypertas->str, qemu_hypertas->len));
1097 g_string_free(qemu_hypertas, TRUE);
1099 _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1100 refpoints, sizeof(refpoints)));
1102 _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1103 maxdomains, sizeof(maxdomains)));
1105 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1106 RTAS_ERROR_LOG_MAX));
1107 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1108 RTAS_EVENT_SCAN_RATE));
1110 g_assert(msi_nonbroken);
1111 _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1114 * According to PAPR, rtas ibm,os-term does not guarantee a return
1115 * back to the guest cpu.
1117 * While an additional ibm,extended-os-term property indicates
1118 * that rtas call return will always occur. Set this property.
1120 _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1122 _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1123 lrdr_capacity, sizeof(lrdr_capacity)));
1125 spapr_dt_rtas_tokens(fdt, rtas);
1129 * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1130 * and the XIVE features that the guest may request and thus the valid
1131 * values for bytes 23..26 of option vector 5:
1133 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1136 PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1139 23, spapr->irq->ov5, /* Xive mode. */
1140 24, 0x00, /* Hash/Radix, filled in below. */
1141 25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1142 26, 0x40, /* Radix options: GTSE == yes. */
1145 if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1146 first_ppc_cpu->compat_pvr)) {
1148 * If we're in a pre POWER9 compat mode then the guest should
1149 * do hash and use the legacy interrupt mode
1151 val[1] = 0x00; /* XICS */
1152 val[3] = 0x00; /* Hash */
1153 } else if (kvm_enabled()) {
1154 if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1155 val[3] = 0x80; /* OV5_MMU_BOTH */
1156 } else if (kvmppc_has_cap_mmu_radix()) {
1157 val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1159 val[3] = 0x00; /* Hash */
1162 /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1165 _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1169 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt)
1171 MachineState *machine = MACHINE(spapr);
1172 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1174 const char *boot_device = machine->boot_order;
1175 char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1177 char *bootlist = get_boot_devices_list(&cb);
1179 _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1181 _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1182 _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1183 spapr->initrd_base));
1184 _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1185 spapr->initrd_base + spapr->initrd_size));
1187 if (spapr->kernel_size) {
1188 uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1189 cpu_to_be64(spapr->kernel_size) };
1191 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1192 &kprop, sizeof(kprop)));
1193 if (spapr->kernel_le) {
1194 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1198 _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1200 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1201 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1202 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1204 if (cb && bootlist) {
1207 for (i = 0; i < cb; i++) {
1208 if (bootlist[i] == '\n') {
1212 _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1215 if (boot_device && strlen(boot_device)) {
1216 _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1219 if (!spapr->has_graphics && stdout_path) {
1221 * "linux,stdout-path" and "stdout" properties are deprecated by linux
1222 * kernel. New platforms should only use the "stdout-path" property. Set
1223 * the new property and continue using older property to remain
1224 * compatible with the existing firmware.
1226 _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1227 _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1230 /* We can deal with BAR reallocation just fine, advertise it to the guest */
1231 if (smc->linux_pci_probe) {
1232 _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1235 spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1237 g_free(stdout_path);
1241 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1243 /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1244 * KVM to work under pHyp with some guest co-operation */
1246 uint8_t hypercall[16];
1248 _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1249 /* indicate KVM hypercall interface */
1250 _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1251 if (kvmppc_has_cap_fixup_hcalls()) {
1253 * Older KVM versions with older guest kernels were broken
1254 * with the magic page, don't allow the guest to map it.
1256 if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1257 sizeof(hypercall))) {
1258 _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1259 hypercall, sizeof(hypercall)));
1264 static void *spapr_build_fdt(SpaprMachineState *spapr)
1266 MachineState *machine = MACHINE(spapr);
1267 MachineClass *mc = MACHINE_GET_CLASS(machine);
1268 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1274 fdt = g_malloc0(FDT_MAX_SIZE);
1275 _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1278 _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1279 _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1280 _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1282 /* Guest UUID & Name*/
1283 buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1284 _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1285 if (qemu_uuid_set) {
1286 _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1290 if (qemu_get_vm_name()) {
1291 _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1292 qemu_get_vm_name()));
1295 /* Host Model & Serial Number */
1296 if (spapr->host_model) {
1297 _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1298 } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1299 _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1303 if (spapr->host_serial) {
1304 _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1305 } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1306 _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1310 _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1311 _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1313 /* /interrupt controller */
1314 spapr->irq->dt_populate(spapr, spapr_max_server_number(spapr), fdt,
1317 ret = spapr_populate_memory(spapr, fdt);
1319 error_report("couldn't setup memory nodes in fdt");
1324 spapr_dt_vdevice(spapr->vio_bus, fdt);
1326 if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1327 ret = spapr_rng_populate_dt(fdt);
1329 error_report("could not set up rng device in the fdt");
1334 QLIST_FOREACH(phb, &spapr->phbs, list) {
1335 ret = spapr_dt_phb(phb, PHANDLE_INTC, fdt, spapr->irq->nr_msis, NULL);
1337 error_report("couldn't setup PCI devices in fdt");
1343 spapr_populate_cpus_dt_node(fdt, spapr);
1345 if (smc->dr_lmb_enabled) {
1346 _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1349 if (mc->has_hotpluggable_cpus) {
1350 int offset = fdt_path_offset(fdt, "/cpus");
1351 ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1353 error_report("Couldn't set up CPU DR device tree properties");
1358 /* /event-sources */
1359 spapr_dt_events(spapr, fdt);
1362 spapr_dt_rtas(spapr, fdt);
1365 spapr_dt_chosen(spapr, fdt);
1368 if (kvm_enabled()) {
1369 spapr_dt_hypervisor(spapr, fdt);
1372 /* Build memory reserve map */
1373 if (spapr->kernel_size) {
1374 _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1376 if (spapr->initrd_size) {
1377 _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1380 /* ibm,client-architecture-support updates */
1381 ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1383 error_report("couldn't setup CAS properties fdt");
1387 if (smc->dr_phb_enabled) {
1388 ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1390 error_report("Couldn't set up PHB DR device tree properties");
1398 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1400 return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1403 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1406 CPUPPCState *env = &cpu->env;
1408 /* The TCG path should also be holding the BQL at this point */
1409 g_assert(qemu_mutex_iothread_locked());
1412 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1413 env->gpr[3] = H_PRIVILEGE;
1415 env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1419 struct LPCRSyncState {
1424 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1426 struct LPCRSyncState *s = arg.host_ptr;
1427 PowerPCCPU *cpu = POWERPC_CPU(cs);
1428 CPUPPCState *env = &cpu->env;
1431 cpu_synchronize_state(cs);
1432 lpcr = env->spr[SPR_LPCR];
1435 ppc_store_lpcr(cpu, lpcr);
1438 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1441 struct LPCRSyncState s = {
1446 run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1450 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1452 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1454 /* Copy PATE1:GR into PATE0:HR */
1455 entry->dw0 = spapr->patb_entry & PATE0_HR;
1456 entry->dw1 = spapr->patb_entry;
1459 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1460 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1461 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1462 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1463 #define DIRTY_HPTE(_hpte) ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1466 * Get the fd to access the kernel htab, re-opening it if necessary
1468 static int get_htab_fd(SpaprMachineState *spapr)
1470 Error *local_err = NULL;
1472 if (spapr->htab_fd >= 0) {
1473 return spapr->htab_fd;
1476 spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1477 if (spapr->htab_fd < 0) {
1478 error_report_err(local_err);
1481 return spapr->htab_fd;
1484 void close_htab_fd(SpaprMachineState *spapr)
1486 if (spapr->htab_fd >= 0) {
1487 close(spapr->htab_fd);
1489 spapr->htab_fd = -1;
1492 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1494 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1496 return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1499 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1501 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1503 assert(kvm_enabled());
1509 return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1512 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1515 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1516 hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1520 * HTAB is controlled by KVM. Fetch into temporary buffer
1522 ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1523 kvmppc_read_hptes(hptes, ptex, n);
1528 * HTAB is controlled by QEMU. Just point to the internally
1531 return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1534 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1535 const ppc_hash_pte64_t *hptes,
1538 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1541 g_free((void *)hptes);
1544 /* Nothing to do for qemu managed HPT */
1547 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1548 uint64_t pte0, uint64_t pte1)
1550 SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1551 hwaddr offset = ptex * HASH_PTE_SIZE_64;
1554 kvmppc_write_hpte(ptex, pte0, pte1);
1556 if (pte0 & HPTE64_V_VALID) {
1557 stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1559 * When setting valid, we write PTE1 first. This ensures
1560 * proper synchronization with the reading code in
1561 * ppc_hash64_pteg_search()
1564 stq_p(spapr->htab + offset, pte0);
1566 stq_p(spapr->htab + offset, pte0);
1568 * When clearing it we set PTE0 first. This ensures proper
1569 * synchronization with the reading code in
1570 * ppc_hash64_pteg_search()
1573 stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1578 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1581 hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1582 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1585 /* There should always be a hash table when this is called */
1586 error_report("spapr_hpte_set_c called with no hash table !");
1590 /* The HW performs a non-atomic byte update */
1591 stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1594 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1597 hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1598 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1601 /* There should always be a hash table when this is called */
1602 error_report("spapr_hpte_set_r called with no hash table !");
1606 /* The HW performs a non-atomic byte update */
1607 stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1610 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1614 /* We aim for a hash table of size 1/128 the size of RAM (rounded
1615 * up). The PAPR recommendation is actually 1/64 of RAM size, but
1616 * that's much more than is needed for Linux guests */
1617 shift = ctz64(pow2ceil(ramsize)) - 7;
1618 shift = MAX(shift, 18); /* Minimum architected size */
1619 shift = MIN(shift, 46); /* Maximum architected size */
1623 void spapr_free_hpt(SpaprMachineState *spapr)
1625 g_free(spapr->htab);
1627 spapr->htab_shift = 0;
1628 close_htab_fd(spapr);
1631 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1636 /* Clean up any HPT info from a previous boot */
1637 spapr_free_hpt(spapr);
1639 rc = kvmppc_reset_htab(shift);
1641 /* kernel-side HPT needed, but couldn't allocate one */
1642 error_setg_errno(errp, errno,
1643 "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1645 /* This is almost certainly fatal, but if the caller really
1646 * wants to carry on with shift == 0, it's welcome to try */
1647 } else if (rc > 0) {
1648 /* kernel-side HPT allocated */
1651 "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1655 spapr->htab_shift = shift;
1658 /* kernel-side HPT not needed, allocate in userspace instead */
1659 size_t size = 1ULL << shift;
1662 spapr->htab = qemu_memalign(size, size);
1664 error_setg_errno(errp, errno,
1665 "Could not allocate HPT of order %d", shift);
1669 memset(spapr->htab, 0, size);
1670 spapr->htab_shift = shift;
1672 for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1673 DIRTY_HPTE(HPTE(spapr->htab, i));
1676 /* We're setting up a hash table, so that means we're not radix */
1677 spapr->patb_entry = 0;
1678 spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1681 void spapr_setup_hpt_and_vrma(SpaprMachineState *spapr)
1685 if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1686 || (spapr->cas_reboot
1687 && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1688 hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1690 uint64_t current_ram_size;
1692 current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1693 hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1695 spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1697 if (spapr->vrma_adjust) {
1698 spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1703 static int spapr_reset_drcs(Object *child, void *opaque)
1706 (SpaprDrc *) object_dynamic_cast(child,
1707 TYPE_SPAPR_DR_CONNECTOR);
1710 spapr_drc_reset(drc);
1716 static void spapr_machine_reset(MachineState *machine)
1718 SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1719 PowerPCCPU *first_ppc_cpu;
1720 uint32_t rtas_limit;
1721 hwaddr rtas_addr, fdt_addr;
1725 spapr_caps_apply(spapr);
1727 first_ppc_cpu = POWERPC_CPU(first_cpu);
1728 if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1729 ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1730 spapr->max_compat_pvr)) {
1732 * If using KVM with radix mode available, VCPUs can be started
1733 * without a HPT because KVM will start them in radix mode.
1734 * Set the GR bit in PATE so that we know there is no HPT.
1736 spapr->patb_entry = PATE1_GR;
1737 spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1739 spapr_setup_hpt_and_vrma(spapr);
1742 qemu_devices_reset();
1745 * If this reset wasn't generated by CAS, we should reset our
1746 * negotiated options and start from scratch
1748 if (!spapr->cas_reboot) {
1749 spapr_ovec_cleanup(spapr->ov5_cas);
1750 spapr->ov5_cas = spapr_ovec_new();
1752 ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1756 * This is fixing some of the default configuration of the XIVE
1757 * devices. To be called after the reset of the machine devices.
1759 spapr_irq_reset(spapr, &error_fatal);
1762 * There is no CAS under qtest. Simulate one to please the code that
1763 * depends on spapr->ov5_cas. This is especially needed to test device
1764 * unplug, so we do that before resetting the DRCs.
1766 if (qtest_enabled()) {
1767 spapr_ovec_cleanup(spapr->ov5_cas);
1768 spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1771 /* DRC reset may cause a device to be unplugged. This will cause troubles
1772 * if this device is used by another device (eg, a running vhost backend
1773 * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1774 * situations, we reset DRCs after all devices have been reset.
1776 object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1778 spapr_clear_pending_events(spapr);
1781 * We place the device tree and RTAS just below either the top of the RMA,
1782 * or just below 2GB, whichever is lower, so that it can be
1783 * processed with 32-bit real mode code if necessary
1785 rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1786 rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1787 fdt_addr = rtas_addr - FDT_MAX_SIZE;
1789 fdt = spapr_build_fdt(spapr);
1791 spapr_load_rtas(spapr, fdt, rtas_addr);
1795 /* Should only fail if we've built a corrupted tree */
1798 if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1799 error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1800 fdt_totalsize(fdt), FDT_MAX_SIZE);
1805 qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1806 cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1807 g_free(spapr->fdt_blob);
1808 spapr->fdt_size = fdt_totalsize(fdt);
1809 spapr->fdt_initial_size = spapr->fdt_size;
1810 spapr->fdt_blob = fdt;
1812 /* Set up the entry state */
1813 spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1814 first_ppc_cpu->env.gpr[5] = 0;
1816 spapr->cas_reboot = false;
1819 static void spapr_create_nvram(SpaprMachineState *spapr)
1821 DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1822 DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1825 qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1829 qdev_init_nofail(dev);
1831 spapr->nvram = (struct SpaprNvram *)dev;
1834 static void spapr_rtc_create(SpaprMachineState *spapr)
1836 object_initialize_child(OBJECT(spapr), "rtc",
1837 &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1838 &error_fatal, NULL);
1839 object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1841 object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1842 "date", &error_fatal);
1845 /* Returns whether we want to use VGA or not */
1846 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1848 switch (vga_interface_type) {
1856 return pci_vga_init(pci_bus) != NULL;
1859 "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1864 static int spapr_pre_load(void *opaque)
1868 rc = spapr_caps_pre_load(opaque);
1876 static int spapr_post_load(void *opaque, int version_id)
1878 SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1881 err = spapr_caps_post_migration(spapr);
1887 * In earlier versions, there was no separate qdev for the PAPR
1888 * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1889 * So when migrating from those versions, poke the incoming offset
1890 * value into the RTC device
1892 if (version_id < 3) {
1893 err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1899 if (kvm_enabled() && spapr->patb_entry) {
1900 PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1901 bool radix = !!(spapr->patb_entry & PATE1_GR);
1902 bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1905 * Update LPCR:HR and UPRT as they may not be set properly in
1908 spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1909 LPCR_HR | LPCR_UPRT);
1911 err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1913 error_report("Process table config unsupported by the host");
1918 err = spapr_irq_post_load(spapr, version_id);
1926 static int spapr_pre_save(void *opaque)
1930 rc = spapr_caps_pre_save(opaque);
1938 static bool version_before_3(void *opaque, int version_id)
1940 return version_id < 3;
1943 static bool spapr_pending_events_needed(void *opaque)
1945 SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1946 return !QTAILQ_EMPTY(&spapr->pending_events);
1949 static const VMStateDescription vmstate_spapr_event_entry = {
1950 .name = "spapr_event_log_entry",
1952 .minimum_version_id = 1,
1953 .fields = (VMStateField[]) {
1954 VMSTATE_UINT32(summary, SpaprEventLogEntry),
1955 VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1956 VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1957 NULL, extended_length),
1958 VMSTATE_END_OF_LIST()
1962 static const VMStateDescription vmstate_spapr_pending_events = {
1963 .name = "spapr_pending_events",
1965 .minimum_version_id = 1,
1966 .needed = spapr_pending_events_needed,
1967 .fields = (VMStateField[]) {
1968 VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1969 vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1970 VMSTATE_END_OF_LIST()
1974 static bool spapr_ov5_cas_needed(void *opaque)
1976 SpaprMachineState *spapr = opaque;
1977 SpaprOptionVector *ov5_mask = spapr_ovec_new();
1978 SpaprOptionVector *ov5_legacy = spapr_ovec_new();
1979 SpaprOptionVector *ov5_removed = spapr_ovec_new();
1982 /* Prior to the introduction of SpaprOptionVector, we had two option
1983 * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1984 * Both of these options encode machine topology into the device-tree
1985 * in such a way that the now-booted OS should still be able to interact
1986 * appropriately with QEMU regardless of what options were actually
1987 * negotiatied on the source side.
1989 * As such, we can avoid migrating the CAS-negotiated options if these
1990 * are the only options available on the current machine/platform.
1991 * Since these are the only options available for pseries-2.7 and
1992 * earlier, this allows us to maintain old->new/new->old migration
1995 * For QEMU 2.8+, there are additional CAS-negotiatable options available
1996 * via default pseries-2.8 machines and explicit command-line parameters.
1997 * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1998 * of the actual CAS-negotiated values to continue working properly. For
1999 * example, availability of memory unplug depends on knowing whether
2000 * OV5_HP_EVT was negotiated via CAS.
2002 * Thus, for any cases where the set of available CAS-negotiatable
2003 * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
2004 * include the CAS-negotiated options in the migration stream, unless
2005 * if they affect boot time behaviour only.
2007 spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
2008 spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
2009 spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
2011 /* spapr_ovec_diff returns true if bits were removed. we avoid using
2012 * the mask itself since in the future it's possible "legacy" bits may be
2013 * removed via machine options, which could generate a false positive
2014 * that breaks migration.
2016 spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
2017 cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
2019 spapr_ovec_cleanup(ov5_mask);
2020 spapr_ovec_cleanup(ov5_legacy);
2021 spapr_ovec_cleanup(ov5_removed);
2026 static const VMStateDescription vmstate_spapr_ov5_cas = {
2027 .name = "spapr_option_vector_ov5_cas",
2029 .minimum_version_id = 1,
2030 .needed = spapr_ov5_cas_needed,
2031 .fields = (VMStateField[]) {
2032 VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
2033 vmstate_spapr_ovec, SpaprOptionVector),
2034 VMSTATE_END_OF_LIST()
2038 static bool spapr_patb_entry_needed(void *opaque)
2040 SpaprMachineState *spapr = opaque;
2042 return !!spapr->patb_entry;
2045 static const VMStateDescription vmstate_spapr_patb_entry = {
2046 .name = "spapr_patb_entry",
2048 .minimum_version_id = 1,
2049 .needed = spapr_patb_entry_needed,
2050 .fields = (VMStateField[]) {
2051 VMSTATE_UINT64(patb_entry, SpaprMachineState),
2052 VMSTATE_END_OF_LIST()
2056 static bool spapr_irq_map_needed(void *opaque)
2058 SpaprMachineState *spapr = opaque;
2060 return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
2063 static const VMStateDescription vmstate_spapr_irq_map = {
2064 .name = "spapr_irq_map",
2066 .minimum_version_id = 1,
2067 .needed = spapr_irq_map_needed,
2068 .fields = (VMStateField[]) {
2069 VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
2070 VMSTATE_END_OF_LIST()
2074 static bool spapr_dtb_needed(void *opaque)
2076 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2078 return smc->update_dt_enabled;
2081 static int spapr_dtb_pre_load(void *opaque)
2083 SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2085 g_free(spapr->fdt_blob);
2086 spapr->fdt_blob = NULL;
2087 spapr->fdt_size = 0;
2092 static const VMStateDescription vmstate_spapr_dtb = {
2093 .name = "spapr_dtb",
2095 .minimum_version_id = 1,
2096 .needed = spapr_dtb_needed,
2097 .pre_load = spapr_dtb_pre_load,
2098 .fields = (VMStateField[]) {
2099 VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2100 VMSTATE_UINT32(fdt_size, SpaprMachineState),
2101 VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2103 VMSTATE_END_OF_LIST()
2107 static const VMStateDescription vmstate_spapr = {
2110 .minimum_version_id = 1,
2111 .pre_load = spapr_pre_load,
2112 .post_load = spapr_post_load,
2113 .pre_save = spapr_pre_save,
2114 .fields = (VMStateField[]) {
2115 /* used to be @next_irq */
2116 VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2119 VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2121 VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2122 VMSTATE_END_OF_LIST()
2124 .subsections = (const VMStateDescription*[]) {
2125 &vmstate_spapr_ov5_cas,
2126 &vmstate_spapr_patb_entry,
2127 &vmstate_spapr_pending_events,
2128 &vmstate_spapr_cap_htm,
2129 &vmstate_spapr_cap_vsx,
2130 &vmstate_spapr_cap_dfp,
2131 &vmstate_spapr_cap_cfpc,
2132 &vmstate_spapr_cap_sbbc,
2133 &vmstate_spapr_cap_ibs,
2134 &vmstate_spapr_cap_hpt_maxpagesize,
2135 &vmstate_spapr_irq_map,
2136 &vmstate_spapr_cap_nested_kvm_hv,
2138 &vmstate_spapr_cap_large_decr,
2139 &vmstate_spapr_cap_ccf_assist,
2144 static int htab_save_setup(QEMUFile *f, void *opaque)
2146 SpaprMachineState *spapr = opaque;
2148 /* "Iteration" header */
2149 if (!spapr->htab_shift) {
2150 qemu_put_be32(f, -1);
2152 qemu_put_be32(f, spapr->htab_shift);
2156 spapr->htab_save_index = 0;
2157 spapr->htab_first_pass = true;
2159 if (spapr->htab_shift) {
2160 assert(kvm_enabled());
2168 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2169 int chunkstart, int n_valid, int n_invalid)
2171 qemu_put_be32(f, chunkstart);
2172 qemu_put_be16(f, n_valid);
2173 qemu_put_be16(f, n_invalid);
2174 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2175 HASH_PTE_SIZE_64 * n_valid);
2178 static void htab_save_end_marker(QEMUFile *f)
2180 qemu_put_be32(f, 0);
2181 qemu_put_be16(f, 0);
2182 qemu_put_be16(f, 0);
2185 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2188 bool has_timeout = max_ns != -1;
2189 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2190 int index = spapr->htab_save_index;
2191 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2193 assert(spapr->htab_first_pass);
2198 /* Consume invalid HPTEs */
2199 while ((index < htabslots)
2200 && !HPTE_VALID(HPTE(spapr->htab, index))) {
2201 CLEAN_HPTE(HPTE(spapr->htab, index));
2205 /* Consume valid HPTEs */
2207 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2208 && HPTE_VALID(HPTE(spapr->htab, index))) {
2209 CLEAN_HPTE(HPTE(spapr->htab, index));
2213 if (index > chunkstart) {
2214 int n_valid = index - chunkstart;
2216 htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2219 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2223 } while ((index < htabslots) && !qemu_file_rate_limit(f));
2225 if (index >= htabslots) {
2226 assert(index == htabslots);
2228 spapr->htab_first_pass = false;
2230 spapr->htab_save_index = index;
2233 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2236 bool final = max_ns < 0;
2237 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2238 int examined = 0, sent = 0;
2239 int index = spapr->htab_save_index;
2240 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2242 assert(!spapr->htab_first_pass);
2245 int chunkstart, invalidstart;
2247 /* Consume non-dirty HPTEs */
2248 while ((index < htabslots)
2249 && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2255 /* Consume valid dirty HPTEs */
2256 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2257 && HPTE_DIRTY(HPTE(spapr->htab, index))
2258 && HPTE_VALID(HPTE(spapr->htab, index))) {
2259 CLEAN_HPTE(HPTE(spapr->htab, index));
2264 invalidstart = index;
2265 /* Consume invalid dirty HPTEs */
2266 while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2267 && HPTE_DIRTY(HPTE(spapr->htab, index))
2268 && !HPTE_VALID(HPTE(spapr->htab, index))) {
2269 CLEAN_HPTE(HPTE(spapr->htab, index));
2274 if (index > chunkstart) {
2275 int n_valid = invalidstart - chunkstart;
2276 int n_invalid = index - invalidstart;
2278 htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2279 sent += index - chunkstart;
2281 if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2286 if (examined >= htabslots) {
2290 if (index >= htabslots) {
2291 assert(index == htabslots);
2294 } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2296 if (index >= htabslots) {
2297 assert(index == htabslots);
2301 spapr->htab_save_index = index;
2303 return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2306 #define MAX_ITERATION_NS 5000000 /* 5 ms */
2307 #define MAX_KVM_BUF_SIZE 2048
2309 static int htab_save_iterate(QEMUFile *f, void *opaque)
2311 SpaprMachineState *spapr = opaque;
2315 /* Iteration header */
2316 if (!spapr->htab_shift) {
2317 qemu_put_be32(f, -1);
2320 qemu_put_be32(f, 0);
2324 assert(kvm_enabled());
2326 fd = get_htab_fd(spapr);
2331 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2335 } else if (spapr->htab_first_pass) {
2336 htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2338 rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2341 htab_save_end_marker(f);
2346 static int htab_save_complete(QEMUFile *f, void *opaque)
2348 SpaprMachineState *spapr = opaque;
2351 /* Iteration header */
2352 if (!spapr->htab_shift) {
2353 qemu_put_be32(f, -1);
2356 qemu_put_be32(f, 0);
2362 assert(kvm_enabled());
2364 fd = get_htab_fd(spapr);
2369 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2374 if (spapr->htab_first_pass) {
2375 htab_save_first_pass(f, spapr, -1);
2377 htab_save_later_pass(f, spapr, -1);
2381 htab_save_end_marker(f);
2386 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2388 SpaprMachineState *spapr = opaque;
2389 uint32_t section_hdr;
2391 Error *local_err = NULL;
2393 if (version_id < 1 || version_id > 1) {
2394 error_report("htab_load() bad version");
2398 section_hdr = qemu_get_be32(f);
2400 if (section_hdr == -1) {
2401 spapr_free_hpt(spapr);
2406 /* First section gives the htab size */
2407 spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2409 error_report_err(local_err);
2416 assert(kvm_enabled());
2418 fd = kvmppc_get_htab_fd(true, 0, &local_err);
2420 error_report_err(local_err);
2427 uint16_t n_valid, n_invalid;
2429 index = qemu_get_be32(f);
2430 n_valid = qemu_get_be16(f);
2431 n_invalid = qemu_get_be16(f);
2433 if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2438 if ((index + n_valid + n_invalid) >
2439 (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2440 /* Bad index in stream */
2442 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2443 index, n_valid, n_invalid, spapr->htab_shift);
2449 qemu_get_buffer(f, HPTE(spapr->htab, index),
2450 HASH_PTE_SIZE_64 * n_valid);
2453 memset(HPTE(spapr->htab, index + n_valid), 0,
2454 HASH_PTE_SIZE_64 * n_invalid);
2461 rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2476 static void htab_save_cleanup(void *opaque)
2478 SpaprMachineState *spapr = opaque;
2480 close_htab_fd(spapr);
2483 static SaveVMHandlers savevm_htab_handlers = {
2484 .save_setup = htab_save_setup,
2485 .save_live_iterate = htab_save_iterate,
2486 .save_live_complete_precopy = htab_save_complete,
2487 .save_cleanup = htab_save_cleanup,
2488 .load_state = htab_load,
2491 static void spapr_boot_set(void *opaque, const char *boot_device,
2494 MachineState *machine = MACHINE(opaque);
2495 machine->boot_order = g_strdup(boot_device);
2498 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2500 MachineState *machine = MACHINE(spapr);
2501 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2502 uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2505 for (i = 0; i < nr_lmbs; i++) {
2508 addr = i * lmb_size + machine->device_memory->base;
2509 spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2515 * If RAM size, maxmem size and individual node mem sizes aren't aligned
2516 * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2517 * since we can't support such unaligned sizes with DRCONF_MEMORY.
2519 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2523 if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2524 error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2525 " is not aligned to %" PRIu64 " MiB",
2527 SPAPR_MEMORY_BLOCK_SIZE / MiB);
2531 if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2532 error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2533 " is not aligned to %" PRIu64 " MiB",
2535 SPAPR_MEMORY_BLOCK_SIZE / MiB);
2539 for (i = 0; i < machine->numa_state->num_nodes; i++) {
2540 if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2542 "Node %d memory size 0x%" PRIx64
2543 " is not aligned to %" PRIu64 " MiB",
2544 i, machine->numa_state->nodes[i].node_mem,
2545 SPAPR_MEMORY_BLOCK_SIZE / MiB);
2551 /* find cpu slot in machine->possible_cpus by core_id */
2552 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2554 int index = id / ms->smp.threads;
2556 if (index >= ms->possible_cpus->len) {
2562 return &ms->possible_cpus->cpus[index];
2565 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2567 MachineState *ms = MACHINE(spapr);
2568 Error *local_err = NULL;
2569 bool vsmt_user = !!spapr->vsmt;
2570 int kvm_smt = kvmppc_smt_threads();
2572 unsigned int smp_threads = ms->smp.threads;
2574 if (!kvm_enabled() && (smp_threads > 1)) {
2575 error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2576 "on a pseries machine");
2579 if (!is_power_of_2(smp_threads)) {
2580 error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2581 "machine because it must be a power of 2", smp_threads);
2585 /* Detemine the VSMT mode to use: */
2587 if (spapr->vsmt < smp_threads) {
2588 error_setg(&local_err, "Cannot support VSMT mode %d"
2589 " because it must be >= threads/core (%d)",
2590 spapr->vsmt, smp_threads);
2593 /* In this case, spapr->vsmt has been set by the command line */
2596 * Default VSMT value is tricky, because we need it to be as
2597 * consistent as possible (for migration), but this requires
2598 * changing it for at least some existing cases. We pick 8 as
2599 * the value that we'd get with KVM on POWER8, the
2600 * overwhelmingly common case in production systems.
2602 spapr->vsmt = MAX(8, smp_threads);
2605 /* KVM: If necessary, set the SMT mode: */
2606 if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2607 ret = kvmppc_set_smt_threads(spapr->vsmt);
2609 /* Looks like KVM isn't able to change VSMT mode */
2610 error_setg(&local_err,
2611 "Failed to set KVM's VSMT mode to %d (errno %d)",
2613 /* We can live with that if the default one is big enough
2614 * for the number of threads, and a submultiple of the one
2615 * we want. In this case we'll waste some vcpu ids, but
2616 * behaviour will be correct */
2617 if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2618 warn_report_err(local_err);
2623 error_append_hint(&local_err,
2624 "On PPC, a VM with %d threads/core"
2625 " on a host with %d threads/core"
2626 " requires the use of VSMT mode %d.\n",
2627 smp_threads, kvm_smt, spapr->vsmt);
2629 kvmppc_hint_smt_possible(&local_err);
2634 /* else TCG: nothing to do currently */
2636 error_propagate(errp, local_err);
2639 static void spapr_init_cpus(SpaprMachineState *spapr)
2641 MachineState *machine = MACHINE(spapr);
2642 MachineClass *mc = MACHINE_GET_CLASS(machine);
2643 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2644 const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2645 const CPUArchIdList *possible_cpus;
2646 unsigned int smp_cpus = machine->smp.cpus;
2647 unsigned int smp_threads = machine->smp.threads;
2648 unsigned int max_cpus = machine->smp.max_cpus;
2649 int boot_cores_nr = smp_cpus / smp_threads;
2652 possible_cpus = mc->possible_cpu_arch_ids(machine);
2653 if (mc->has_hotpluggable_cpus) {
2654 if (smp_cpus % smp_threads) {
2655 error_report("smp_cpus (%u) must be multiple of threads (%u)",
2656 smp_cpus, smp_threads);
2659 if (max_cpus % smp_threads) {
2660 error_report("max_cpus (%u) must be multiple of threads (%u)",
2661 max_cpus, smp_threads);
2665 if (max_cpus != smp_cpus) {
2666 error_report("This machine version does not support CPU hotplug");
2669 boot_cores_nr = possible_cpus->len;
2672 if (smc->pre_2_10_has_unused_icps) {
2675 for (i = 0; i < spapr_max_server_number(spapr); i++) {
2676 /* Dummy entries get deregistered when real ICPState objects
2677 * are registered during CPU core hotplug.
2679 pre_2_10_vmstate_register_dummy_icp(i);
2683 for (i = 0; i < possible_cpus->len; i++) {
2684 int core_id = i * smp_threads;
2686 if (mc->has_hotpluggable_cpus) {
2687 spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2688 spapr_vcpu_id(spapr, core_id));
2691 if (i < boot_cores_nr) {
2692 Object *core = object_new(type);
2693 int nr_threads = smp_threads;
2695 /* Handle the partially filled core for older machine types */
2696 if ((i + 1) * smp_threads >= smp_cpus) {
2697 nr_threads = smp_cpus - i * smp_threads;
2700 object_property_set_int(core, nr_threads, "nr-threads",
2702 object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2704 object_property_set_bool(core, true, "realized", &error_fatal);
2711 static PCIHostState *spapr_create_default_phb(void)
2715 dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2716 qdev_prop_set_uint32(dev, "index", 0);
2717 qdev_init_nofail(dev);
2719 return PCI_HOST_BRIDGE(dev);
2722 /* pSeries LPAR / sPAPR hardware init */
2723 static void spapr_machine_init(MachineState *machine)
2725 SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2726 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2727 const char *kernel_filename = machine->kernel_filename;
2728 const char *initrd_filename = machine->initrd_filename;
2731 MemoryRegion *sysmem = get_system_memory();
2732 MemoryRegion *ram = g_new(MemoryRegion, 1);
2733 hwaddr node0_size = spapr_node0_size(machine);
2734 long load_limit, fw_size;
2736 Error *resize_hpt_err = NULL;
2738 msi_nonbroken = true;
2740 QLIST_INIT(&spapr->phbs);
2741 QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2743 /* Determine capabilities to run with */
2744 spapr_caps_init(spapr);
2746 kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2747 if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2749 * If the user explicitly requested a mode we should either
2750 * supply it, or fail completely (which we do below). But if
2751 * it's not set explicitly, we reset our mode to something
2754 if (resize_hpt_err) {
2755 spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2756 error_free(resize_hpt_err);
2757 resize_hpt_err = NULL;
2759 spapr->resize_hpt = smc->resize_hpt_default;
2763 assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2765 if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2767 * User requested HPT resize, but this host can't supply it. Bail out
2769 error_report_err(resize_hpt_err);
2773 spapr->rma_size = node0_size;
2775 /* With KVM, we don't actually know whether KVM supports an
2776 * unbounded RMA (PR KVM) or is limited by the hash table size
2777 * (HV KVM using VRMA), so we always assume the latter
2779 * In that case, we also limit the initial allocations for RTAS
2780 * etc... to 256M since we have no way to know what the VRMA size
2781 * is going to be as it depends on the size of the hash table
2782 * which isn't determined yet.
2784 if (kvm_enabled()) {
2785 spapr->vrma_adjust = 1;
2786 spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2789 /* Actually we don't support unbounded RMA anymore since we added
2790 * proper emulation of HV mode. The max we can get is 16G which
2791 * also happens to be what we configure for PAPR mode so make sure
2792 * we don't do anything bigger than that
2794 spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2796 if (spapr->rma_size > node0_size) {
2797 error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2802 /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2803 load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2806 * VSMT must be set in order to be able to compute VCPU ids, ie to
2807 * call spapr_max_server_number() or spapr_vcpu_id().
2809 spapr_set_vsmt_mode(spapr, &error_fatal);
2811 /* Set up Interrupt Controller before we create the VCPUs */
2812 spapr_irq_init(spapr, &error_fatal);
2814 /* Set up containers for ibm,client-architecture-support negotiated options
2816 spapr->ov5 = spapr_ovec_new();
2817 spapr->ov5_cas = spapr_ovec_new();
2819 if (smc->dr_lmb_enabled) {
2820 spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2821 spapr_validate_node_memory(machine, &error_fatal);
2824 spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2826 /* advertise support for dedicated HP event source to guests */
2827 if (spapr->use_hotplug_event_source) {
2828 spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2831 /* advertise support for HPT resizing */
2832 if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2833 spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2836 /* advertise support for ibm,dyamic-memory-v2 */
2837 spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2839 /* advertise XIVE on POWER9 machines */
2840 if (spapr->irq->ov5 & (SPAPR_OV5_XIVE_EXPLOIT | SPAPR_OV5_XIVE_BOTH)) {
2841 spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2845 spapr_init_cpus(spapr);
2848 * check we don't have a memory-less/cpu-less NUMA node
2849 * Firmware relies on the existing memory/cpu topology to provide the
2850 * NUMA topology to the kernel.
2851 * And the linux kernel needs to know the NUMA topology at start
2852 * to be able to hotplug CPUs later.
2854 if (machine->numa_state->num_nodes) {
2855 for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2856 /* check for memory-less node */
2857 if (machine->numa_state->nodes[i].node_mem == 0) {
2860 /* check for cpu-less node */
2862 PowerPCCPU *cpu = POWERPC_CPU(cs);
2863 if (cpu->node_id == i) {
2868 /* memory-less and cpu-less node */
2871 "Memory-less/cpu-less nodes are not supported (node %d)",
2881 * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2882 * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2883 * called from vPHB reset handler so we initialize the counter here.
2884 * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2885 * must be equally distant from any other node.
2886 * The final value of spapr->gpu_numa_id is going to be written to
2887 * max-associativity-domains in spapr_build_fdt().
2889 spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2891 if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2892 ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2893 spapr->max_compat_pvr)) {
2894 /* KVM and TCG always allow GTSE with radix... */
2895 spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2897 /* ... but not with hash (currently). */
2899 if (kvm_enabled()) {
2900 /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2901 kvmppc_enable_logical_ci_hcalls();
2902 kvmppc_enable_set_mode_hcall();
2904 /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2905 kvmppc_enable_clear_ref_mod_hcalls();
2907 /* Enable H_PAGE_INIT */
2908 kvmppc_enable_h_page_init();
2912 memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2914 memory_region_add_subregion(sysmem, 0, ram);
2916 /* always allocate the device memory information */
2917 machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2919 /* initialize hotplug memory address space */
2920 if (machine->ram_size < machine->maxram_size) {
2921 ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2923 * Limit the number of hotpluggable memory slots to half the number
2924 * slots that KVM supports, leaving the other half for PCI and other
2925 * devices. However ensure that number of slots doesn't drop below 32.
2927 int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2928 SPAPR_MAX_RAM_SLOTS;
2930 if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2931 max_memslots = SPAPR_MAX_RAM_SLOTS;
2933 if (machine->ram_slots > max_memslots) {
2934 error_report("Specified number of memory slots %"
2935 PRIu64" exceeds max supported %d",
2936 machine->ram_slots, max_memslots);
2940 machine->device_memory->base = ROUND_UP(machine->ram_size,
2941 SPAPR_DEVICE_MEM_ALIGN);
2942 memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2943 "device-memory", device_mem_size);
2944 memory_region_add_subregion(sysmem, machine->device_memory->base,
2945 &machine->device_memory->mr);
2948 if (smc->dr_lmb_enabled) {
2949 spapr_create_lmb_dr_connectors(spapr);
2952 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2954 error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2957 spapr->rtas_size = get_image_size(filename);
2958 if (spapr->rtas_size < 0) {
2959 error_report("Could not get size of LPAR rtas '%s'", filename);
2962 spapr->rtas_blob = g_malloc(spapr->rtas_size);
2963 if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2964 error_report("Could not load LPAR rtas '%s'", filename);
2967 if (spapr->rtas_size > RTAS_MAX_SIZE) {
2968 error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2969 (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2974 /* Set up RTAS event infrastructure */
2975 spapr_events_init(spapr);
2977 /* Set up the RTC RTAS interfaces */
2978 spapr_rtc_create(spapr);
2980 /* Set up VIO bus */
2981 spapr->vio_bus = spapr_vio_bus_init();
2983 for (i = 0; i < serial_max_hds(); i++) {
2985 spapr_vty_create(spapr->vio_bus, serial_hd(i));
2989 /* We always have at least the nvram device on VIO */
2990 spapr_create_nvram(spapr);
2993 * Setup hotplug / dynamic-reconfiguration connectors. top-level
2994 * connectors (described in root DT node's "ibm,drc-types" property)
2995 * are pre-initialized here. additional child connectors (such as
2996 * connectors for a PHBs PCI slots) are added as needed during their
2997 * parent's realization.
2999 if (smc->dr_phb_enabled) {
3000 for (i = 0; i < SPAPR_MAX_PHBS; i++) {
3001 spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
3006 spapr_pci_rtas_init();
3008 phb = spapr_create_default_phb();
3010 for (i = 0; i < nb_nics; i++) {
3011 NICInfo *nd = &nd_table[i];
3014 nd->model = g_strdup("spapr-vlan");
3017 if (g_str_equal(nd->model, "spapr-vlan") ||
3018 g_str_equal(nd->model, "ibmveth")) {
3019 spapr_vlan_create(spapr->vio_bus, nd);
3021 pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
3025 for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
3026 spapr_vscsi_create(spapr->vio_bus);
3030 if (spapr_vga_init(phb->bus, &error_fatal)) {
3031 spapr->has_graphics = true;
3032 machine->usb |= defaults_enabled() && !machine->usb_disabled;
3036 if (smc->use_ohci_by_default) {
3037 pci_create_simple(phb->bus, -1, "pci-ohci");
3039 pci_create_simple(phb->bus, -1, "nec-usb-xhci");
3042 if (spapr->has_graphics) {
3043 USBBus *usb_bus = usb_bus_find(-1);
3045 usb_create_simple(usb_bus, "usb-kbd");
3046 usb_create_simple(usb_bus, "usb-mouse");
3050 if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
3052 "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
3057 if (kernel_filename) {
3058 uint64_t lowaddr = 0;
3060 spapr->kernel_size = load_elf(kernel_filename, NULL,
3061 translate_kernel_address, NULL,
3062 NULL, &lowaddr, NULL, 1,
3063 PPC_ELF_MACHINE, 0, 0);
3064 if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
3065 spapr->kernel_size = load_elf(kernel_filename, NULL,
3066 translate_kernel_address, NULL, NULL,
3067 &lowaddr, NULL, 0, PPC_ELF_MACHINE,
3069 spapr->kernel_le = spapr->kernel_size > 0;
3071 if (spapr->kernel_size < 0) {
3072 error_report("error loading %s: %s", kernel_filename,
3073 load_elf_strerror(spapr->kernel_size));
3078 if (initrd_filename) {
3079 /* Try to locate the initrd in the gap between the kernel
3080 * and the firmware. Add a bit of space just in case
3082 spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
3083 + 0x1ffff) & ~0xffff;
3084 spapr->initrd_size = load_image_targphys(initrd_filename,
3087 - spapr->initrd_base);
3088 if (spapr->initrd_size < 0) {
3089 error_report("could not load initial ram disk '%s'",
3096 if (bios_name == NULL) {
3097 bios_name = FW_FILE_NAME;
3099 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3101 error_report("Could not find LPAR firmware '%s'", bios_name);
3104 fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3106 error_report("Could not load LPAR firmware '%s'", filename);
3111 /* FIXME: Should register things through the MachineState's qdev
3112 * interface, this is a legacy from the sPAPREnvironment structure
3113 * which predated MachineState but had a similar function */
3114 vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3115 register_savevm_live("spapr/htab", -1, 1,
3116 &savevm_htab_handlers, spapr);
3118 qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3121 qemu_register_boot_set(spapr_boot_set, spapr);
3124 * Nothing needs to be done to resume a suspended guest because
3125 * suspending does not change the machine state, so no need for
3126 * a ->wakeup method.
3128 qemu_register_wakeup_support();
3130 if (kvm_enabled()) {
3131 /* to stop and start vmclock */
3132 qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3135 kvmppc_spapr_enable_inkernel_multitce();
3139 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3145 if (!strcmp(vm_type, "HV")) {
3149 if (!strcmp(vm_type, "PR")) {
3153 error_report("Unknown kvm-type specified '%s'", vm_type);
3158 * Implementation of an interface to adjust firmware path
3159 * for the bootindex property handling.
3161 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3164 #define CAST(type, obj, name) \
3165 ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3166 SCSIDevice *d = CAST(SCSIDevice, dev, TYPE_SCSI_DEVICE);
3167 SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3168 VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3171 void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3172 VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3173 USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3177 * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3178 * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3179 * 0x8000 | (target << 8) | (bus << 5) | lun
3180 * (see the "Logical unit addressing format" table in SAM5)
3182 unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3183 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3184 (uint64_t)id << 48);
3185 } else if (virtio) {
3187 * We use SRP luns of the form 01000000 | (target << 8) | lun
3188 * in the top 32 bits of the 64-bit LUN
3189 * Note: the quote above is from SLOF and it is wrong,
3190 * the actual binding is:
3191 * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3193 unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3194 if (d->lun >= 256) {
3195 /* Use the LUN "flat space addressing method" */
3198 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3199 (uint64_t)id << 32);
3202 * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3203 * in the top 32 bits of the 64-bit LUN
3205 unsigned usb_port = atoi(usb->port->path);
3206 unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3207 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3208 (uint64_t)id << 32);
3213 * SLOF probes the USB devices, and if it recognizes that the device is a
3214 * storage device, it changes its name to "storage" instead of "usb-host",
3215 * and additionally adds a child node for the SCSI LUN, so the correct
3216 * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3218 if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3219 USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3220 if (usb_host_dev_is_scsi_storage(usbdev)) {
3221 return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3226 /* Replace "pci" with "pci@800000020000000" */
3227 return g_strdup_printf("pci@%"PRIX64, phb->buid);
3231 /* Same logic as virtio above */
3232 unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3233 return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3236 if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3237 /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3238 PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3239 return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3245 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3247 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3249 return g_strdup(spapr->kvm_type);
3252 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3254 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3256 g_free(spapr->kvm_type);
3257 spapr->kvm_type = g_strdup(value);
3260 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3262 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3264 return spapr->use_hotplug_event_source;
3267 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3270 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3272 spapr->use_hotplug_event_source = value;
3275 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3280 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3282 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3284 switch (spapr->resize_hpt) {
3285 case SPAPR_RESIZE_HPT_DEFAULT:
3286 return g_strdup("default");
3287 case SPAPR_RESIZE_HPT_DISABLED:
3288 return g_strdup("disabled");
3289 case SPAPR_RESIZE_HPT_ENABLED:
3290 return g_strdup("enabled");
3291 case SPAPR_RESIZE_HPT_REQUIRED:
3292 return g_strdup("required");
3294 g_assert_not_reached();
3297 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3299 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3301 if (strcmp(value, "default") == 0) {
3302 spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3303 } else if (strcmp(value, "disabled") == 0) {
3304 spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3305 } else if (strcmp(value, "enabled") == 0) {
3306 spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3307 } else if (strcmp(value, "required") == 0) {
3308 spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3310 error_setg(errp, "Bad value for \"resize-hpt\" property");
3314 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3315 void *opaque, Error **errp)
3317 visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3320 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3321 void *opaque, Error **errp)
3323 visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3326 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3328 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3330 if (spapr->irq == &spapr_irq_xics_legacy) {
3331 return g_strdup("legacy");
3332 } else if (spapr->irq == &spapr_irq_xics) {
3333 return g_strdup("xics");
3334 } else if (spapr->irq == &spapr_irq_xive) {
3335 return g_strdup("xive");
3336 } else if (spapr->irq == &spapr_irq_dual) {
3337 return g_strdup("dual");
3339 g_assert_not_reached();
3342 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3344 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3346 if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3347 error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3351 /* The legacy IRQ backend can not be set */
3352 if (strcmp(value, "xics") == 0) {
3353 spapr->irq = &spapr_irq_xics;
3354 } else if (strcmp(value, "xive") == 0) {
3355 spapr->irq = &spapr_irq_xive;
3356 } else if (strcmp(value, "dual") == 0) {
3357 spapr->irq = &spapr_irq_dual;
3359 error_setg(errp, "Bad value for \"ic-mode\" property");
3363 static char *spapr_get_host_model(Object *obj, Error **errp)
3365 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3367 return g_strdup(spapr->host_model);
3370 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3372 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3374 g_free(spapr->host_model);
3375 spapr->host_model = g_strdup(value);
3378 static char *spapr_get_host_serial(Object *obj, Error **errp)
3380 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3382 return g_strdup(spapr->host_serial);
3385 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3387 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3389 g_free(spapr->host_serial);
3390 spapr->host_serial = g_strdup(value);
3393 static void spapr_instance_init(Object *obj)
3395 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3396 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3398 spapr->htab_fd = -1;
3399 spapr->use_hotplug_event_source = true;
3400 object_property_add_str(obj, "kvm-type",
3401 spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3402 object_property_set_description(obj, "kvm-type",
3403 "Specifies the KVM virtualization mode (HV, PR)",
3405 object_property_add_bool(obj, "modern-hotplug-events",
3406 spapr_get_modern_hotplug_events,
3407 spapr_set_modern_hotplug_events,
3409 object_property_set_description(obj, "modern-hotplug-events",
3410 "Use dedicated hotplug event mechanism in"
3411 " place of standard EPOW events when possible"
3412 " (required for memory hot-unplug support)",
3414 ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3415 "Maximum permitted CPU compatibility mode",
3418 object_property_add_str(obj, "resize-hpt",
3419 spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3420 object_property_set_description(obj, "resize-hpt",
3421 "Resizing of the Hash Page Table (enabled, disabled, required)",
3423 object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3424 spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3425 object_property_set_description(obj, "vsmt",
3426 "Virtual SMT: KVM behaves as if this were"
3427 " the host's SMT mode", &error_abort);
3428 object_property_add_bool(obj, "vfio-no-msix-emulation",
3429 spapr_get_msix_emulation, NULL, NULL);
3431 /* The machine class defines the default interrupt controller mode */
3432 spapr->irq = smc->irq;
3433 object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3434 spapr_set_ic_mode, NULL);
3435 object_property_set_description(obj, "ic-mode",
3436 "Specifies the interrupt controller mode (xics, xive, dual)",
3439 object_property_add_str(obj, "host-model",
3440 spapr_get_host_model, spapr_set_host_model,
3442 object_property_set_description(obj, "host-model",
3443 "Host model to advertise in guest device tree", &error_abort);
3444 object_property_add_str(obj, "host-serial",
3445 spapr_get_host_serial, spapr_set_host_serial,
3447 object_property_set_description(obj, "host-serial",
3448 "Host serial number to advertise in guest device tree", &error_abort);
3451 static void spapr_machine_finalizefn(Object *obj)
3453 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3455 g_free(spapr->kvm_type);
3458 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3460 cpu_synchronize_state(cs);
3461 ppc_cpu_do_system_reset(cs);
3464 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3469 async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3473 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3474 void *fdt, int *fdt_start_offset, Error **errp)
3479 addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3480 node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3482 *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3483 SPAPR_MEMORY_BLOCK_SIZE);
3487 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3488 bool dedicated_hp_event_source, Error **errp)
3491 uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3493 uint64_t addr = addr_start;
3494 bool hotplugged = spapr_drc_hotplugged(dev);
3495 Error *local_err = NULL;
3497 for (i = 0; i < nr_lmbs; i++) {
3498 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3499 addr / SPAPR_MEMORY_BLOCK_SIZE);
3502 spapr_drc_attach(drc, dev, &local_err);
3504 while (addr > addr_start) {
3505 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3506 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3507 addr / SPAPR_MEMORY_BLOCK_SIZE);
3508 spapr_drc_detach(drc);
3510 error_propagate(errp, local_err);
3514 spapr_drc_reset(drc);
3516 addr += SPAPR_MEMORY_BLOCK_SIZE;
3518 /* send hotplug notification to the
3519 * guest only in case of hotplugged memory
3522 if (dedicated_hp_event_source) {
3523 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3524 addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3525 spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3527 spapr_drc_index(drc));
3529 spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3535 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3538 Error *local_err = NULL;
3539 SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3540 PCDIMMDevice *dimm = PC_DIMM(dev);
3541 uint64_t size, addr;
3543 size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3545 pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3550 addr = object_property_get_uint(OBJECT(dimm),
3551 PC_DIMM_ADDR_PROP, &local_err);
3556 spapr_add_lmbs(dev, addr, size, spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3565 pc_dimm_unplug(dimm, MACHINE(ms));
3567 error_propagate(errp, local_err);
3570 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3573 const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3574 SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3575 PCDIMMDevice *dimm = PC_DIMM(dev);
3576 Error *local_err = NULL;
3581 if (!smc->dr_lmb_enabled) {
3582 error_setg(errp, "Memory hotplug not supported for this machine");
3586 size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3588 error_propagate(errp, local_err);
3592 if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3593 error_setg(errp, "Hotplugged memory size must be a multiple of "
3594 "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3598 memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3600 pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3601 spapr_check_pagesize(spapr, pagesize, &local_err);
3603 error_propagate(errp, local_err);
3607 pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3610 struct SpaprDimmState {
3613 QTAILQ_ENTRY(SpaprDimmState) next;
3616 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3619 SpaprDimmState *dimm_state = NULL;
3621 QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3622 if (dimm_state->dimm == dimm) {
3629 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3633 SpaprDimmState *ds = NULL;
3636 * If this request is for a DIMM whose removal had failed earlier
3637 * (due to guest's refusal to remove the LMBs), we would have this
3638 * dimm already in the pending_dimm_unplugs list. In that
3639 * case don't add again.
3641 ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3643 ds = g_malloc0(sizeof(SpaprDimmState));
3644 ds->nr_lmbs = nr_lmbs;
3646 QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3651 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3652 SpaprDimmState *dimm_state)
3654 QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3658 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3662 uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3664 uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3665 uint32_t avail_lmbs = 0;
3666 uint64_t addr_start, addr;
3669 addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3673 for (i = 0; i < nr_lmbs; i++) {
3674 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3675 addr / SPAPR_MEMORY_BLOCK_SIZE);
3680 addr += SPAPR_MEMORY_BLOCK_SIZE;
3683 return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3686 /* Callback to be called during DRC release. */
3687 void spapr_lmb_release(DeviceState *dev)
3689 HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3690 SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3691 SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3693 /* This information will get lost if a migration occurs
3694 * during the unplug process. In this case recover it. */
3696 ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3698 /* The DRC being examined by the caller at least must be counted */
3699 g_assert(ds->nr_lmbs);
3702 if (--ds->nr_lmbs) {
3707 * Now that all the LMBs have been removed by the guest, call the
3708 * unplug handler chain. This can never fail.
3710 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3711 object_unparent(OBJECT(dev));
3714 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3716 SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3717 SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3719 pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3720 object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3721 spapr_pending_dimm_unplugs_remove(spapr, ds);
3724 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3725 DeviceState *dev, Error **errp)
3727 SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3728 Error *local_err = NULL;
3729 PCDIMMDevice *dimm = PC_DIMM(dev);
3731 uint64_t size, addr_start, addr;
3735 size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3736 nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3738 addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3745 * An existing pending dimm state for this DIMM means that there is an
3746 * unplug operation in progress, waiting for the spapr_lmb_release
3747 * callback to complete the job (BQL can't cover that far). In this case,
3748 * bail out to avoid detaching DRCs that were already released.
3750 if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3751 error_setg(&local_err,
3752 "Memory unplug already in progress for device %s",
3757 spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3760 for (i = 0; i < nr_lmbs; i++) {
3761 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3762 addr / SPAPR_MEMORY_BLOCK_SIZE);
3765 spapr_drc_detach(drc);
3766 addr += SPAPR_MEMORY_BLOCK_SIZE;
3769 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3770 addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3771 spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3772 nr_lmbs, spapr_drc_index(drc));
3774 error_propagate(errp, local_err);
3777 /* Callback to be called during DRC release. */
3778 void spapr_core_release(DeviceState *dev)
3780 HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3782 /* Call the unplug handler chain. This can never fail. */
3783 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3784 object_unparent(OBJECT(dev));
3787 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3789 MachineState *ms = MACHINE(hotplug_dev);
3790 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3791 CPUCore *cc = CPU_CORE(dev);
3792 CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3794 if (smc->pre_2_10_has_unused_icps) {
3795 SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3798 for (i = 0; i < cc->nr_threads; i++) {
3799 CPUState *cs = CPU(sc->threads[i]);
3801 pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3806 core_slot->cpu = NULL;
3807 object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3811 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3814 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3817 CPUCore *cc = CPU_CORE(dev);
3819 if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3820 error_setg(errp, "Unable to find CPU core with core-id: %d",
3825 error_setg(errp, "Boot CPU core may not be unplugged");
3829 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3830 spapr_vcpu_id(spapr, cc->core_id));
3833 spapr_drc_detach(drc);
3835 spapr_hotplug_req_remove_by_index(drc);
3838 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3839 void *fdt, int *fdt_start_offset, Error **errp)
3841 SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3842 CPUState *cs = CPU(core->threads[0]);
3843 PowerPCCPU *cpu = POWERPC_CPU(cs);
3844 DeviceClass *dc = DEVICE_GET_CLASS(cs);
3845 int id = spapr_get_vcpu_id(cpu);
3849 nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3850 offset = fdt_add_subnode(fdt, 0, nodename);
3853 spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3855 *fdt_start_offset = offset;
3859 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3862 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3863 MachineClass *mc = MACHINE_GET_CLASS(spapr);
3864 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3865 SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3866 CPUCore *cc = CPU_CORE(dev);
3869 Error *local_err = NULL;
3870 CPUArchId *core_slot;
3872 bool hotplugged = spapr_drc_hotplugged(dev);
3875 core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3877 error_setg(errp, "Unable to find CPU core with core-id: %d",
3881 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3882 spapr_vcpu_id(spapr, cc->core_id));
3884 g_assert(drc || !mc->has_hotpluggable_cpus);
3887 spapr_drc_attach(drc, dev, &local_err);
3889 error_propagate(errp, local_err);
3895 * Send hotplug notification interrupt to the guest only
3896 * in case of hotplugged CPUs.
3898 spapr_hotplug_req_add_by_index(drc);
3900 spapr_drc_reset(drc);
3904 core_slot->cpu = OBJECT(dev);
3906 if (smc->pre_2_10_has_unused_icps) {
3907 for (i = 0; i < cc->nr_threads; i++) {
3908 cs = CPU(core->threads[i]);
3909 pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3914 * Set compatibility mode to match the boot CPU, which was either set
3915 * by the machine reset code or by CAS.
3918 for (i = 0; i < cc->nr_threads; i++) {
3919 ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3922 error_propagate(errp, local_err);
3929 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3932 MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3933 MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3934 Error *local_err = NULL;
3935 CPUCore *cc = CPU_CORE(dev);
3936 const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3937 const char *type = object_get_typename(OBJECT(dev));
3938 CPUArchId *core_slot;
3940 unsigned int smp_threads = machine->smp.threads;
3942 if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3943 error_setg(&local_err, "CPU hotplug not supported for this machine");
3947 if (strcmp(base_core_type, type)) {
3948 error_setg(&local_err, "CPU core type should be %s", base_core_type);
3952 if (cc->core_id % smp_threads) {
3953 error_setg(&local_err, "invalid core id %d", cc->core_id);
3958 * In general we should have homogeneous threads-per-core, but old
3959 * (pre hotplug support) machine types allow the last core to have
3960 * reduced threads as a compatibility hack for when we allowed
3961 * total vcpus not a multiple of threads-per-core.
3963 if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3964 error_setg(&local_err, "invalid nr-threads %d, must be %d",
3965 cc->nr_threads, smp_threads);
3969 core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3971 error_setg(&local_err, "core id %d out of range", cc->core_id);
3975 if (core_slot->cpu) {
3976 error_setg(&local_err, "core %d already populated", cc->core_id);
3980 numa_cpu_pre_plug(core_slot, dev, &local_err);
3983 error_propagate(errp, local_err);
3986 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3987 void *fdt, int *fdt_start_offset, Error **errp)
3989 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3992 intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3993 if (intc_phandle <= 0) {
3997 if (spapr_dt_phb(sphb, intc_phandle, fdt, spapr->irq->nr_msis,
3998 fdt_start_offset)) {
3999 error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4003 /* generally SLOF creates these, for hotplug it's up to QEMU */
4004 _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4009 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4012 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4013 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4014 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4015 const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4017 if (dev->hotplugged && !smc->dr_phb_enabled) {
4018 error_setg(errp, "PHB hotplug not supported for this machine");
4022 if (sphb->index == (uint32_t)-1) {
4023 error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4028 * This will check that sphb->index doesn't exceed the maximum number of
4029 * PHBs for the current machine type.
4031 smc->phb_placement(spapr, sphb->index,
4032 &sphb->buid, &sphb->io_win_addr,
4033 &sphb->mem_win_addr, &sphb->mem64_win_addr,
4034 windows_supported, sphb->dma_liobn,
4035 &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4039 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4042 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4043 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4044 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4046 bool hotplugged = spapr_drc_hotplugged(dev);
4047 Error *local_err = NULL;
4049 if (!smc->dr_phb_enabled) {
4053 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4054 /* hotplug hooks should check it's enabled before getting this far */
4057 spapr_drc_attach(drc, DEVICE(dev), &local_err);
4059 error_propagate(errp, local_err);
4064 spapr_hotplug_req_add_by_index(drc);
4066 spapr_drc_reset(drc);
4070 void spapr_phb_release(DeviceState *dev)
4072 HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4074 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4075 object_unparent(OBJECT(dev));
4078 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4080 object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4083 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4084 DeviceState *dev, Error **errp)
4086 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4089 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4092 if (!spapr_drc_unplug_requested(drc)) {
4093 spapr_drc_detach(drc);
4094 spapr_hotplug_req_remove_by_index(drc);
4098 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4101 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4102 SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4104 if (spapr->tpm_proxy != NULL) {
4105 error_setg(errp, "Only one TPM proxy can be specified for this machine");
4109 spapr->tpm_proxy = tpm_proxy;
4112 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4114 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4116 object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4117 object_unparent(OBJECT(dev));
4118 spapr->tpm_proxy = NULL;
4121 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4122 DeviceState *dev, Error **errp)
4124 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4125 spapr_memory_plug(hotplug_dev, dev, errp);
4126 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4127 spapr_core_plug(hotplug_dev, dev, errp);
4128 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4129 spapr_phb_plug(hotplug_dev, dev, errp);
4130 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4131 spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4135 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4136 DeviceState *dev, Error **errp)
4138 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4139 spapr_memory_unplug(hotplug_dev, dev);
4140 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4141 spapr_core_unplug(hotplug_dev, dev);
4142 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4143 spapr_phb_unplug(hotplug_dev, dev);
4144 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4145 spapr_tpm_proxy_unplug(hotplug_dev, dev);
4149 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4150 DeviceState *dev, Error **errp)
4152 SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4153 MachineClass *mc = MACHINE_GET_CLASS(sms);
4154 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4156 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4157 if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4158 spapr_memory_unplug_request(hotplug_dev, dev, errp);
4160 /* NOTE: this means there is a window after guest reset, prior to
4161 * CAS negotiation, where unplug requests will fail due to the
4162 * capability not being detected yet. This is a bit different than
4163 * the case with PCI unplug, where the events will be queued and
4164 * eventually handled by the guest after boot
4166 error_setg(errp, "Memory hot unplug not supported for this guest");
4168 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4169 if (!mc->has_hotpluggable_cpus) {
4170 error_setg(errp, "CPU hot unplug not supported on this machine");
4173 spapr_core_unplug_request(hotplug_dev, dev, errp);
4174 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4175 if (!smc->dr_phb_enabled) {
4176 error_setg(errp, "PHB hot unplug not supported on this machine");
4179 spapr_phb_unplug_request(hotplug_dev, dev, errp);
4180 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4181 spapr_tpm_proxy_unplug(hotplug_dev, dev);
4185 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4186 DeviceState *dev, Error **errp)
4188 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4189 spapr_memory_pre_plug(hotplug_dev, dev, errp);
4190 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4191 spapr_core_pre_plug(hotplug_dev, dev, errp);
4192 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4193 spapr_phb_pre_plug(hotplug_dev, dev, errp);
4197 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4200 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4201 object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4202 object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4203 object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4204 return HOTPLUG_HANDLER(machine);
4206 if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4207 PCIDevice *pcidev = PCI_DEVICE(dev);
4208 PCIBus *root = pci_device_root_bus(pcidev);
4209 SpaprPhbState *phb =
4210 (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4211 TYPE_SPAPR_PCI_HOST_BRIDGE);
4214 return HOTPLUG_HANDLER(phb);
4220 static CpuInstanceProperties
4221 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4223 CPUArchId *core_slot;
4224 MachineClass *mc = MACHINE_GET_CLASS(machine);
4226 /* make sure possible_cpu are intialized */
4227 mc->possible_cpu_arch_ids(machine);
4228 /* get CPU core slot containing thread that matches cpu_index */
4229 core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4231 return core_slot->props;
4234 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4236 return idx / ms->smp.cores % ms->numa_state->num_nodes;
4239 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4242 unsigned int smp_threads = machine->smp.threads;
4243 unsigned int smp_cpus = machine->smp.cpus;
4244 const char *core_type;
4245 int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4246 MachineClass *mc = MACHINE_GET_CLASS(machine);
4248 if (!mc->has_hotpluggable_cpus) {
4249 spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4251 if (machine->possible_cpus) {
4252 assert(machine->possible_cpus->len == spapr_max_cores);
4253 return machine->possible_cpus;
4256 core_type = spapr_get_cpu_core_type(machine->cpu_type);
4258 error_report("Unable to find sPAPR CPU Core definition");
4262 machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4263 sizeof(CPUArchId) * spapr_max_cores);
4264 machine->possible_cpus->len = spapr_max_cores;
4265 for (i = 0; i < machine->possible_cpus->len; i++) {
4266 int core_id = i * smp_threads;
4268 machine->possible_cpus->cpus[i].type = core_type;
4269 machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4270 machine->possible_cpus->cpus[i].arch_id = core_id;
4271 machine->possible_cpus->cpus[i].props.has_core_id = true;
4272 machine->possible_cpus->cpus[i].props.core_id = core_id;
4274 return machine->possible_cpus;
4277 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4278 uint64_t *buid, hwaddr *pio,
4279 hwaddr *mmio32, hwaddr *mmio64,
4280 unsigned n_dma, uint32_t *liobns,
4281 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4284 * New-style PHB window placement.
4286 * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4287 * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4290 * Some guest kernels can't work with MMIO windows above 1<<46
4291 * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4293 * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4294 * PHB stacked together. (32TiB+2GiB)..(32TiB+64GiB) contains the
4295 * 2GiB 32-bit MMIO windows for each PHB. Then 33..64TiB has the
4296 * 1TiB 64-bit MMIO windows for each PHB.
4298 const uint64_t base_buid = 0x800000020000000ULL;
4301 /* Sanity check natural alignments */
4302 QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4303 QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4304 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4305 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4306 /* Sanity check bounds */
4307 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4308 SPAPR_PCI_MEM32_WIN_SIZE);
4309 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4310 SPAPR_PCI_MEM64_WIN_SIZE);
4312 if (index >= SPAPR_MAX_PHBS) {
4313 error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4314 SPAPR_MAX_PHBS - 1);
4318 *buid = base_buid + index;
4319 for (i = 0; i < n_dma; ++i) {
4320 liobns[i] = SPAPR_PCI_LIOBN(index, i);
4323 *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4324 *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4325 *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4327 *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4328 *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4331 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4333 SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4335 return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4338 static void spapr_ics_resend(XICSFabric *dev)
4340 SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4342 ics_resend(spapr->ics);
4345 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4347 PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4349 return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4352 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4355 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4357 spapr->irq->print_info(spapr, mon);
4358 monitor_printf(mon, "irqchip: %s\n",
4359 kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4362 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4364 return cpu->vcpu_id;
4367 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4369 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4370 MachineState *ms = MACHINE(spapr);
4373 vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4375 if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4376 error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4377 error_append_hint(errp, "Adjust the number of cpus to %d "
4378 "or try to raise the number of threads per core\n",
4379 vcpu_id * ms->smp.threads / spapr->vsmt);
4383 cpu->vcpu_id = vcpu_id;
4386 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4391 PowerPCCPU *cpu = POWERPC_CPU(cs);
4393 if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4401 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4403 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4405 /* These are only called by TCG, KVM maintains dispatch state */
4407 spapr_cpu->prod = false;
4408 if (spapr_cpu->vpa_addr) {
4409 CPUState *cs = CPU(cpu);
4412 dispatch = ldl_be_phys(cs->as,
4413 spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4415 if ((dispatch & 1) != 0) {
4416 qemu_log_mask(LOG_GUEST_ERROR,
4417 "VPA: incorrect dispatch counter value for "
4418 "dispatched partition %u, correcting.\n", dispatch);
4422 spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4426 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4428 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4430 if (spapr_cpu->vpa_addr) {
4431 CPUState *cs = CPU(cpu);
4434 dispatch = ldl_be_phys(cs->as,
4435 spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4437 if ((dispatch & 1) != 1) {
4438 qemu_log_mask(LOG_GUEST_ERROR,
4439 "VPA: incorrect dispatch counter value for "
4440 "preempted partition %u, correcting.\n", dispatch);
4444 spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4448 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4450 MachineClass *mc = MACHINE_CLASS(oc);
4451 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4452 FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4453 NMIClass *nc = NMI_CLASS(oc);
4454 HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4455 PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4456 XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4457 InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4459 mc->desc = "pSeries Logical Partition (PAPR compliant)";
4460 mc->ignore_boot_device_suffixes = true;
4463 * We set up the default / latest behaviour here. The class_init
4464 * functions for the specific versioned machine types can override
4465 * these details for backwards compatibility
4467 mc->init = spapr_machine_init;
4468 mc->reset = spapr_machine_reset;
4469 mc->block_default_type = IF_SCSI;
4470 mc->max_cpus = 1024;
4471 mc->no_parallel = 1;
4472 mc->default_boot_order = "";
4473 mc->default_ram_size = 512 * MiB;
4474 mc->default_display = "std";
4475 mc->kvm_type = spapr_kvm_type;
4476 machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4477 mc->pci_allow_0_address = true;
4478 assert(!mc->get_hotplug_handler);
4479 mc->get_hotplug_handler = spapr_get_hotplug_handler;
4480 hc->pre_plug = spapr_machine_device_pre_plug;
4481 hc->plug = spapr_machine_device_plug;
4482 mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4483 mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4484 mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4485 hc->unplug_request = spapr_machine_device_unplug_request;
4486 hc->unplug = spapr_machine_device_unplug;
4488 smc->dr_lmb_enabled = true;
4489 smc->update_dt_enabled = true;
4490 mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4491 mc->has_hotpluggable_cpus = true;
4492 smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4493 fwc->get_dev_path = spapr_get_fw_dev_path;
4494 nc->nmi_monitor_handler = spapr_nmi;
4495 smc->phb_placement = spapr_phb_placement;
4496 vhc->hypercall = emulate_spapr_hypercall;
4497 vhc->hpt_mask = spapr_hpt_mask;
4498 vhc->map_hptes = spapr_map_hptes;
4499 vhc->unmap_hptes = spapr_unmap_hptes;
4500 vhc->hpte_set_c = spapr_hpte_set_c;
4501 vhc->hpte_set_r = spapr_hpte_set_r;
4502 vhc->get_pate = spapr_get_pate;
4503 vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4504 vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4505 vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4506 xic->ics_get = spapr_ics_get;
4507 xic->ics_resend = spapr_ics_resend;
4508 xic->icp_get = spapr_icp_get;
4509 ispc->print_info = spapr_pic_print_info;
4510 /* Force NUMA node memory size to be a multiple of
4511 * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4512 * in which LMBs are represented and hot-added
4514 mc->numa_mem_align_shift = 28;
4515 mc->numa_mem_supported = true;
4517 smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4518 smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4519 smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4520 smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4521 smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4522 smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4523 smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4524 smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4525 smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4526 smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4527 spapr_caps_add_properties(smc, &error_abort);
4528 smc->irq = &spapr_irq_dual;
4529 smc->dr_phb_enabled = true;
4530 smc->linux_pci_probe = true;
4533 static const TypeInfo spapr_machine_info = {
4534 .name = TYPE_SPAPR_MACHINE,
4535 .parent = TYPE_MACHINE,
4537 .instance_size = sizeof(SpaprMachineState),
4538 .instance_init = spapr_instance_init,
4539 .instance_finalize = spapr_machine_finalizefn,
4540 .class_size = sizeof(SpaprMachineClass),
4541 .class_init = spapr_machine_class_init,
4542 .interfaces = (InterfaceInfo[]) {
4543 { TYPE_FW_PATH_PROVIDER },
4545 { TYPE_HOTPLUG_HANDLER },
4546 { TYPE_PPC_VIRTUAL_HYPERVISOR },
4547 { TYPE_XICS_FABRIC },
4548 { TYPE_INTERRUPT_STATS_PROVIDER },
4553 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest) \
4554 static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4557 MachineClass *mc = MACHINE_CLASS(oc); \
4558 spapr_machine_##suffix##_class_options(mc); \
4560 mc->alias = "pseries"; \
4561 mc->is_default = 1; \
4564 static const TypeInfo spapr_machine_##suffix##_info = { \
4565 .name = MACHINE_TYPE_NAME("pseries-" verstr), \
4566 .parent = TYPE_SPAPR_MACHINE, \
4567 .class_init = spapr_machine_##suffix##_class_init, \
4569 static void spapr_machine_register_##suffix(void) \
4571 type_register(&spapr_machine_##suffix##_info); \
4573 type_init(spapr_machine_register_##suffix)
4578 static void spapr_machine_4_2_class_options(MachineClass *mc)
4580 /* Defaults for the latest behaviour inherited from the base class */
4583 DEFINE_SPAPR_MACHINE(4_2, "4.2", true);
4588 static void spapr_machine_4_1_class_options(MachineClass *mc)
4590 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4591 static GlobalProperty compat[] = {
4592 /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4593 { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4596 spapr_machine_4_2_class_options(mc);
4597 smc->linux_pci_probe = false;
4598 compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4599 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4602 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4607 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4608 uint64_t *buid, hwaddr *pio,
4609 hwaddr *mmio32, hwaddr *mmio64,
4610 unsigned n_dma, uint32_t *liobns,
4611 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4613 spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4614 nv2gpa, nv2atsd, errp);
4619 static void spapr_machine_4_0_class_options(MachineClass *mc)
4621 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4623 spapr_machine_4_1_class_options(mc);
4624 compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4625 smc->phb_placement = phb_placement_4_0;
4626 smc->irq = &spapr_irq_xics;
4627 smc->pre_4_1_migration = true;
4630 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4635 static void spapr_machine_3_1_class_options(MachineClass *mc)
4637 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4639 spapr_machine_4_0_class_options(mc);
4640 compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4642 mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4643 smc->update_dt_enabled = false;
4644 smc->dr_phb_enabled = false;
4645 smc->broken_host_serial_model = true;
4646 smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4647 smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4648 smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4649 smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4652 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4658 static void spapr_machine_3_0_class_options(MachineClass *mc)
4660 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4662 spapr_machine_3_1_class_options(mc);
4663 compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4665 smc->legacy_irq_allocation = true;
4666 smc->irq = &spapr_irq_xics_legacy;
4669 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4674 static void spapr_machine_2_12_class_options(MachineClass *mc)
4676 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4677 static GlobalProperty compat[] = {
4678 { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4679 { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4682 spapr_machine_3_0_class_options(mc);
4683 compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4684 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4686 /* We depend on kvm_enabled() to choose a default value for the
4687 * hpt-max-page-size capability. Of course we can't do it here
4688 * because this is too early and the HW accelerator isn't initialzed
4689 * yet. Postpone this to machine init (see default_caps_with_cpu()).
4691 smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4694 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4696 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4698 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4700 spapr_machine_2_12_class_options(mc);
4701 smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4702 smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4703 smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4706 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4712 static void spapr_machine_2_11_class_options(MachineClass *mc)
4714 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4716 spapr_machine_2_12_class_options(mc);
4717 smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4718 compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4721 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4727 static void spapr_machine_2_10_class_options(MachineClass *mc)
4729 spapr_machine_2_11_class_options(mc);
4730 compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4733 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4739 static void spapr_machine_2_9_class_options(MachineClass *mc)
4741 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4742 static GlobalProperty compat[] = {
4743 { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4746 spapr_machine_2_10_class_options(mc);
4747 compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4748 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4749 mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4750 smc->pre_2_10_has_unused_icps = true;
4751 smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4754 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4760 static void spapr_machine_2_8_class_options(MachineClass *mc)
4762 static GlobalProperty compat[] = {
4763 { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4766 spapr_machine_2_9_class_options(mc);
4767 compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4768 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4769 mc->numa_mem_align_shift = 23;
4772 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4778 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4779 uint64_t *buid, hwaddr *pio,
4780 hwaddr *mmio32, hwaddr *mmio64,
4781 unsigned n_dma, uint32_t *liobns,
4782 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4784 /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4785 const uint64_t base_buid = 0x800000020000000ULL;
4786 const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4787 const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4788 const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4789 const uint32_t max_index = 255;
4790 const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4792 uint64_t ram_top = MACHINE(spapr)->ram_size;
4793 hwaddr phb0_base, phb_base;
4796 /* Do we have device memory? */
4797 if (MACHINE(spapr)->maxram_size > ram_top) {
4798 /* Can't just use maxram_size, because there may be an
4799 * alignment gap between normal and device memory regions
4801 ram_top = MACHINE(spapr)->device_memory->base +
4802 memory_region_size(&MACHINE(spapr)->device_memory->mr);
4805 phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4807 if (index > max_index) {
4808 error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4813 *buid = base_buid + index;
4814 for (i = 0; i < n_dma; ++i) {
4815 liobns[i] = SPAPR_PCI_LIOBN(index, i);
4818 phb_base = phb0_base + index * phb_spacing;
4819 *pio = phb_base + pio_offset;
4820 *mmio32 = phb_base + mmio_offset;
4822 * We don't set the 64-bit MMIO window, relying on the PHB's
4823 * fallback behaviour of automatically splitting a large "32-bit"
4824 * window into contiguous 32-bit and 64-bit windows
4831 static void spapr_machine_2_7_class_options(MachineClass *mc)
4833 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4834 static GlobalProperty compat[] = {
4835 { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4836 { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4837 { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4838 { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4841 spapr_machine_2_8_class_options(mc);
4842 mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4843 mc->default_machine_opts = "modern-hotplug-events=off";
4844 compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4845 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4846 smc->phb_placement = phb_placement_2_7;
4849 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4855 static void spapr_machine_2_6_class_options(MachineClass *mc)
4857 static GlobalProperty compat[] = {
4858 { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4861 spapr_machine_2_7_class_options(mc);
4862 mc->has_hotpluggable_cpus = false;
4863 compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4864 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4867 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4873 static void spapr_machine_2_5_class_options(MachineClass *mc)
4875 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4876 static GlobalProperty compat[] = {
4877 { "spapr-vlan", "use-rx-buffer-pools", "off" },
4880 spapr_machine_2_6_class_options(mc);
4881 smc->use_ohci_by_default = true;
4882 compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4883 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4886 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4892 static void spapr_machine_2_4_class_options(MachineClass *mc)
4894 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4896 spapr_machine_2_5_class_options(mc);
4897 smc->dr_lmb_enabled = false;
4898 compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4901 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4907 static void spapr_machine_2_3_class_options(MachineClass *mc)
4909 static GlobalProperty compat[] = {
4910 { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4912 spapr_machine_2_4_class_options(mc);
4913 compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4914 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4916 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4922 static void spapr_machine_2_2_class_options(MachineClass *mc)
4924 static GlobalProperty compat[] = {
4925 { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4928 spapr_machine_2_3_class_options(mc);
4929 compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4930 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4931 mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4933 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4939 static void spapr_machine_2_1_class_options(MachineClass *mc)
4941 spapr_machine_2_2_class_options(mc);
4942 compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4944 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4946 static void spapr_machine_register_types(void)
4948 type_register_static(&spapr_machine_info);
4951 type_init(spapr_machine_register_types)