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 "qapi/error.h"
29 #include "sysemu/sysemu.h"
30 #include "sysemu/numa.h"
33 #include "hw/fw-path-provider.h"
36 #include "sysemu/device_tree.h"
37 #include "sysemu/block-backend.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
41 #include "migration/migration.h"
42 #include "mmu-hash64.h"
45 #include "hw/boards.h"
46 #include "hw/ppc/ppc.h"
47 #include "hw/loader.h"
49 #include "hw/ppc/fdt.h"
50 #include "hw/ppc/spapr.h"
51 #include "hw/ppc/spapr_vio.h"
52 #include "hw/pci-host/spapr.h"
53 #include "hw/ppc/xics.h"
54 #include "hw/pci/msi.h"
56 #include "hw/pci/pci.h"
57 #include "hw/scsi/scsi.h"
58 #include "hw/virtio/virtio-scsi.h"
60 #include "exec/address-spaces.h"
62 #include "qemu/config-file.h"
63 #include "qemu/error-report.h"
66 #include "hw/intc/intc.h"
68 #include "hw/compat.h"
69 #include "qemu/cutils.h"
70 #include "hw/ppc/spapr_cpu_core.h"
71 #include "qmp-commands.h"
75 /* SLOF memory layout:
77 * SLOF raw image loaded at 0, copies its romfs right below the flat
78 * device-tree, then position SLOF itself 31M below that
80 * So we set FW_OVERHEAD to 40MB which should account for all of that
83 * We load our kernel at 4M, leaving space for SLOF initial image
85 #define FDT_MAX_SIZE 0x100000
86 #define RTAS_MAX_SIZE 0x10000
87 #define RTAS_MAX_ADDR 0x80000000 /* RTAS must stay below that */
88 #define FW_MAX_SIZE 0x400000
89 #define FW_FILE_NAME "slof.bin"
90 #define FW_OVERHEAD 0x2800000
91 #define KERNEL_LOAD_ADDR FW_MAX_SIZE
93 #define MIN_RMA_SLOF 128UL
95 #define PHANDLE_XICP 0x00001111
97 #define HTAB_SIZE(spapr) (1ULL << ((spapr)->htab_shift))
99 static int try_create_xics(sPAPRMachineState *spapr, const char *type_ics,
100 const char *type_icp, int nr_servers,
101 int nr_irqs, Error **errp)
103 XICSFabric *xi = XICS_FABRIC(spapr);
104 Error *err = NULL, *local_err = NULL;
105 ICSState *ics = NULL;
108 ics = ICS_SIMPLE(object_new(type_ics));
109 qdev_set_parent_bus(DEVICE(ics), sysbus_get_default());
110 object_property_add_child(OBJECT(spapr), "ics", OBJECT(ics), NULL);
111 object_property_set_int(OBJECT(ics), nr_irqs, "nr-irqs", &err);
112 object_property_add_const_link(OBJECT(ics), "xics", OBJECT(xi), NULL);
113 object_property_set_bool(OBJECT(ics), true, "realized", &local_err);
114 error_propagate(&err, local_err);
119 spapr->icps = g_malloc0(nr_servers * sizeof(ICPState));
120 spapr->nr_servers = nr_servers;
122 for (i = 0; i < nr_servers; i++) {
123 ICPState *icp = &spapr->icps[i];
125 object_initialize(icp, sizeof(*icp), type_icp);
126 qdev_set_parent_bus(DEVICE(icp), sysbus_get_default());
127 object_property_add_child(OBJECT(spapr), "icp[*]", OBJECT(icp), NULL);
128 object_property_add_const_link(OBJECT(icp), "xics", OBJECT(xi), NULL);
129 object_property_set_bool(OBJECT(icp), true, "realized", &err);
133 object_unref(OBJECT(icp));
140 error_propagate(errp, err);
142 object_unparent(OBJECT(ics));
147 static int xics_system_init(MachineState *machine,
148 int nr_servers, int nr_irqs, Error **errp)
155 if (machine_kernel_irqchip_allowed(machine) &&
156 !xics_kvm_init(SPAPR_MACHINE(machine), errp)) {
157 rc = try_create_xics(SPAPR_MACHINE(machine), TYPE_ICS_KVM,
158 TYPE_KVM_ICP, nr_servers, nr_irqs, &err);
160 if (machine_kernel_irqchip_required(machine) && rc < 0) {
161 error_reportf_err(err,
162 "kernel_irqchip requested but unavailable: ");
169 xics_spapr_init(SPAPR_MACHINE(machine), errp);
170 rc = try_create_xics(SPAPR_MACHINE(machine), TYPE_ICS_SIMPLE,
171 TYPE_ICP, nr_servers, nr_irqs, errp);
177 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
181 uint32_t servers_prop[smt_threads];
182 uint32_t gservers_prop[smt_threads * 2];
183 int index = ppc_get_vcpu_dt_id(cpu);
185 if (cpu->compat_pvr) {
186 ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
192 /* Build interrupt servers and gservers properties */
193 for (i = 0; i < smt_threads; i++) {
194 servers_prop[i] = cpu_to_be32(index + i);
195 /* Hack, direct the group queues back to cpu 0 */
196 gservers_prop[i*2] = cpu_to_be32(index + i);
197 gservers_prop[i*2 + 1] = 0;
199 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
200 servers_prop, sizeof(servers_prop));
204 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
205 gservers_prop, sizeof(gservers_prop));
210 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
213 PowerPCCPU *cpu = POWERPC_CPU(cs);
214 int index = ppc_get_vcpu_dt_id(cpu);
215 uint32_t associativity[] = {cpu_to_be32(0x5),
219 cpu_to_be32(cs->numa_node),
222 /* Advertise NUMA via ibm,associativity */
223 if (nb_numa_nodes > 1) {
224 ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
225 sizeof(associativity));
231 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
233 int ret = 0, offset, cpus_offset;
236 int smt = kvmppc_smt_threads();
237 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
240 PowerPCCPU *cpu = POWERPC_CPU(cs);
241 DeviceClass *dc = DEVICE_GET_CLASS(cs);
242 int index = ppc_get_vcpu_dt_id(cpu);
243 int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
245 if ((index % smt) != 0) {
249 snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
251 cpus_offset = fdt_path_offset(fdt, "/cpus");
252 if (cpus_offset < 0) {
253 cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
255 if (cpus_offset < 0) {
259 offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
261 offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
267 ret = fdt_setprop(fdt, offset, "ibm,pft-size",
268 pft_size_prop, sizeof(pft_size_prop));
273 ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
278 ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
286 static hwaddr spapr_node0_size(void)
288 MachineState *machine = MACHINE(qdev_get_machine());
292 for (i = 0; i < nb_numa_nodes; ++i) {
293 if (numa_info[i].node_mem) {
294 return MIN(pow2floor(numa_info[i].node_mem),
299 return machine->ram_size;
302 static void add_str(GString *s, const gchar *s1)
304 g_string_append_len(s, s1, strlen(s1) + 1);
307 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
310 uint32_t associativity[] = {
311 cpu_to_be32(0x4), /* length */
312 cpu_to_be32(0x0), cpu_to_be32(0x0),
313 cpu_to_be32(0x0), cpu_to_be32(nodeid)
316 uint64_t mem_reg_property[2];
319 mem_reg_property[0] = cpu_to_be64(start);
320 mem_reg_property[1] = cpu_to_be64(size);
322 sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
323 off = fdt_add_subnode(fdt, 0, mem_name);
325 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
326 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
327 sizeof(mem_reg_property))));
328 _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
329 sizeof(associativity))));
333 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
335 MachineState *machine = MACHINE(spapr);
336 hwaddr mem_start, node_size;
337 int i, nb_nodes = nb_numa_nodes;
338 NodeInfo *nodes = numa_info;
341 /* No NUMA nodes, assume there is just one node with whole RAM */
342 if (!nb_numa_nodes) {
344 ramnode.node_mem = machine->ram_size;
348 for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
349 if (!nodes[i].node_mem) {
352 if (mem_start >= machine->ram_size) {
355 node_size = nodes[i].node_mem;
356 if (node_size > machine->ram_size - mem_start) {
357 node_size = machine->ram_size - mem_start;
361 /* ppc_spapr_init() checks for rma_size <= node0_size already */
362 spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
363 mem_start += spapr->rma_size;
364 node_size -= spapr->rma_size;
366 for ( ; node_size; ) {
367 hwaddr sizetmp = pow2floor(node_size);
369 /* mem_start != 0 here */
370 if (ctzl(mem_start) < ctzl(sizetmp)) {
371 sizetmp = 1ULL << ctzl(mem_start);
374 spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
375 node_size -= sizetmp;
376 mem_start += sizetmp;
383 /* Populate the "ibm,pa-features" property */
384 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset)
386 uint8_t pa_features_206[] = { 6, 0,
387 0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
388 uint8_t pa_features_207[] = { 24, 0,
389 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
390 0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
391 0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
392 0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
393 uint8_t *pa_features;
396 switch (env->mmu_model) {
397 case POWERPC_MMU_2_06:
398 case POWERPC_MMU_2_06a:
399 pa_features = pa_features_206;
400 pa_size = sizeof(pa_features_206);
402 case POWERPC_MMU_2_07:
403 case POWERPC_MMU_2_07a:
404 pa_features = pa_features_207;
405 pa_size = sizeof(pa_features_207);
411 if (env->ci_large_pages) {
413 * Note: we keep CI large pages off by default because a 64K capable
414 * guest provisioned with large pages might otherwise try to map a qemu
415 * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
416 * even if that qemu runs on a 4k host.
417 * We dd this bit back here if we are confident this is not an issue
419 pa_features[3] |= 0x20;
421 if (kvmppc_has_cap_htm() && pa_size > 24) {
422 pa_features[24] |= 0x80; /* Transactional memory support */
425 _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
428 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
429 sPAPRMachineState *spapr)
431 PowerPCCPU *cpu = POWERPC_CPU(cs);
432 CPUPPCState *env = &cpu->env;
433 PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
434 int index = ppc_get_vcpu_dt_id(cpu);
435 uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
436 0xffffffff, 0xffffffff};
437 uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
438 : SPAPR_TIMEBASE_FREQ;
439 uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
440 uint32_t page_sizes_prop[64];
441 size_t page_sizes_prop_size;
442 uint32_t vcpus_per_socket = smp_threads * smp_cores;
443 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
444 int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
445 sPAPRDRConnector *drc;
446 sPAPRDRConnectorClass *drck;
449 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
451 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
452 drc_index = drck->get_index(drc);
453 _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
456 _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
457 _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
459 _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
460 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
461 env->dcache_line_size)));
462 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
463 env->dcache_line_size)));
464 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
465 env->icache_line_size)));
466 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
467 env->icache_line_size)));
469 if (pcc->l1_dcache_size) {
470 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
471 pcc->l1_dcache_size)));
473 error_report("Warning: Unknown L1 dcache size for cpu");
475 if (pcc->l1_icache_size) {
476 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
477 pcc->l1_icache_size)));
479 error_report("Warning: Unknown L1 icache size for cpu");
482 _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
483 _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
484 _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
485 _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
486 _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
487 _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
489 if (env->spr_cb[SPR_PURR].oea_read) {
490 _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
493 if (env->mmu_model & POWERPC_MMU_1TSEG) {
494 _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
495 segs, sizeof(segs))));
498 /* Advertise VMX/VSX (vector extensions) if available
499 * 0 / no property == no vector extensions
500 * 1 == VMX / Altivec available
501 * 2 == VSX available */
502 if (env->insns_flags & PPC_ALTIVEC) {
503 uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
505 _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
508 /* Advertise DFP (Decimal Floating Point) if available
509 * 0 / no property == no DFP
510 * 1 == DFP available */
511 if (env->insns_flags2 & PPC2_DFP) {
512 _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
515 page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
516 sizeof(page_sizes_prop));
517 if (page_sizes_prop_size) {
518 _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
519 page_sizes_prop, page_sizes_prop_size)));
522 spapr_populate_pa_features(env, fdt, offset);
524 _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
525 cs->cpu_index / vcpus_per_socket)));
527 _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
528 pft_size_prop, sizeof(pft_size_prop))));
530 _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
532 _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
535 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
540 int smt = kvmppc_smt_threads();
542 cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
544 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
545 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
548 * We walk the CPUs in reverse order to ensure that CPU DT nodes
549 * created by fdt_add_subnode() end up in the right order in FDT
550 * for the guest kernel the enumerate the CPUs correctly.
552 CPU_FOREACH_REVERSE(cs) {
553 PowerPCCPU *cpu = POWERPC_CPU(cs);
554 int index = ppc_get_vcpu_dt_id(cpu);
555 DeviceClass *dc = DEVICE_GET_CLASS(cs);
558 if ((index % smt) != 0) {
562 nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
563 offset = fdt_add_subnode(fdt, cpus_offset, nodename);
566 spapr_populate_cpu_dt(cs, fdt, offset, spapr);
572 * Adds ibm,dynamic-reconfiguration-memory node.
573 * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
574 * of this device tree node.
576 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
578 MachineState *machine = MACHINE(spapr);
580 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
581 uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
582 uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
583 uint32_t nr_lmbs = (spapr->hotplug_memory.base +
584 memory_region_size(&spapr->hotplug_memory.mr)) /
586 uint32_t *int_buf, *cur_index, buf_len;
587 int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
590 * Don't create the node if there is no hotpluggable memory
592 if (machine->ram_size == machine->maxram_size) {
597 * Allocate enough buffer size to fit in ibm,dynamic-memory
598 * or ibm,associativity-lookup-arrays
600 buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
602 cur_index = int_buf = g_malloc0(buf_len);
604 offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
606 ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
607 sizeof(prop_lmb_size));
612 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
617 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
622 /* ibm,dynamic-memory */
623 int_buf[0] = cpu_to_be32(nr_lmbs);
625 for (i = 0; i < nr_lmbs; i++) {
626 uint64_t addr = i * lmb_size;
627 uint32_t *dynamic_memory = cur_index;
629 if (i >= hotplug_lmb_start) {
630 sPAPRDRConnector *drc;
631 sPAPRDRConnectorClass *drck;
633 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
635 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
637 dynamic_memory[0] = cpu_to_be32(addr >> 32);
638 dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
639 dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
640 dynamic_memory[3] = cpu_to_be32(0); /* reserved */
641 dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
642 if (memory_region_present(get_system_memory(), addr)) {
643 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
645 dynamic_memory[5] = cpu_to_be32(0);
649 * LMB information for RMA, boot time RAM and gap b/n RAM and
650 * hotplug memory region -- all these are marked as reserved
651 * and as having no valid DRC.
653 dynamic_memory[0] = cpu_to_be32(addr >> 32);
654 dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
655 dynamic_memory[2] = cpu_to_be32(0);
656 dynamic_memory[3] = cpu_to_be32(0); /* reserved */
657 dynamic_memory[4] = cpu_to_be32(-1);
658 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
659 SPAPR_LMB_FLAGS_DRC_INVALID);
662 cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
664 ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
669 /* ibm,associativity-lookup-arrays */
671 int_buf[0] = cpu_to_be32(nr_nodes);
672 int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
674 for (i = 0; i < nr_nodes; i++) {
675 uint32_t associativity[] = {
681 memcpy(cur_index, associativity, sizeof(associativity));
684 ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
685 (cur_index - int_buf) * sizeof(uint32_t));
691 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
692 sPAPROptionVector *ov5_updates)
694 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
697 /* Generate ibm,dynamic-reconfiguration-memory node if required */
698 if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
699 g_assert(smc->dr_lmb_enabled);
700 ret = spapr_populate_drconf_memory(spapr, fdt);
706 offset = fdt_path_offset(fdt, "/chosen");
708 offset = fdt_add_subnode(fdt, 0, "chosen");
713 ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
714 "ibm,architecture-vec-5");
720 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
721 target_ulong addr, target_ulong size,
722 sPAPROptionVector *ov5_updates)
724 void *fdt, *fdt_skel;
725 sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
729 /* Create sceleton */
730 fdt_skel = g_malloc0(size);
731 _FDT((fdt_create(fdt_skel, size)));
732 _FDT((fdt_begin_node(fdt_skel, "")));
733 _FDT((fdt_end_node(fdt_skel)));
734 _FDT((fdt_finish(fdt_skel)));
735 fdt = g_malloc0(size);
736 _FDT((fdt_open_into(fdt_skel, fdt, size)));
739 /* Fixup cpu nodes */
740 _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
742 if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
746 /* Pack resulting tree */
747 _FDT((fdt_pack(fdt)));
749 if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
750 trace_spapr_cas_failed(size);
754 cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
755 cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
756 trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
762 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
765 GString *hypertas = g_string_sized_new(256);
766 GString *qemu_hypertas = g_string_sized_new(256);
767 uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
768 uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
769 memory_region_size(&spapr->hotplug_memory.mr);
770 uint32_t lrdr_capacity[] = {
771 cpu_to_be32(max_hotplug_addr >> 32),
772 cpu_to_be32(max_hotplug_addr & 0xffffffff),
773 0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
774 cpu_to_be32(max_cpus / smp_threads),
777 _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
780 add_str(hypertas, "hcall-pft");
781 add_str(hypertas, "hcall-term");
782 add_str(hypertas, "hcall-dabr");
783 add_str(hypertas, "hcall-interrupt");
784 add_str(hypertas, "hcall-tce");
785 add_str(hypertas, "hcall-vio");
786 add_str(hypertas, "hcall-splpar");
787 add_str(hypertas, "hcall-bulk");
788 add_str(hypertas, "hcall-set-mode");
789 add_str(hypertas, "hcall-sprg0");
790 add_str(hypertas, "hcall-copy");
791 add_str(hypertas, "hcall-debug");
792 add_str(qemu_hypertas, "hcall-memop1");
794 if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
795 add_str(hypertas, "hcall-multi-tce");
797 _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
798 hypertas->str, hypertas->len));
799 g_string_free(hypertas, TRUE);
800 _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
801 qemu_hypertas->str, qemu_hypertas->len));
802 g_string_free(qemu_hypertas, TRUE);
804 _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
805 refpoints, sizeof(refpoints)));
807 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
808 RTAS_ERROR_LOG_MAX));
809 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
810 RTAS_EVENT_SCAN_RATE));
813 _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
817 * According to PAPR, rtas ibm,os-term does not guarantee a return
818 * back to the guest cpu.
820 * While an additional ibm,extended-os-term property indicates
821 * that rtas call return will always occur. Set this property.
823 _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
825 _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
826 lrdr_capacity, sizeof(lrdr_capacity)));
828 spapr_dt_rtas_tokens(fdt, rtas);
831 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
833 MachineState *machine = MACHINE(spapr);
835 const char *boot_device = machine->boot_order;
836 char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
838 char *bootlist = get_boot_devices_list(&cb, true);
840 _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
842 _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
843 _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
844 spapr->initrd_base));
845 _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
846 spapr->initrd_base + spapr->initrd_size));
848 if (spapr->kernel_size) {
849 uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
850 cpu_to_be64(spapr->kernel_size) };
852 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
853 &kprop, sizeof(kprop)));
854 if (spapr->kernel_le) {
855 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
859 _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
861 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
862 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
863 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
865 if (cb && bootlist) {
868 for (i = 0; i < cb; i++) {
869 if (bootlist[i] == '\n') {
873 _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
876 if (boot_device && strlen(boot_device)) {
877 _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
880 if (!spapr->has_graphics && stdout_path) {
881 _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
888 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
890 /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
891 * KVM to work under pHyp with some guest co-operation */
893 uint8_t hypercall[16];
895 _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
896 /* indicate KVM hypercall interface */
897 _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
898 if (kvmppc_has_cap_fixup_hcalls()) {
900 * Older KVM versions with older guest kernels were broken
901 * with the magic page, don't allow the guest to map it.
903 if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
904 sizeof(hypercall))) {
905 _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
906 hypercall, sizeof(hypercall)));
911 static void *spapr_build_fdt(sPAPRMachineState *spapr,
915 MachineState *machine = MACHINE(qdev_get_machine());
916 MachineClass *mc = MACHINE_GET_CLASS(machine);
917 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
923 fdt = g_malloc0(FDT_MAX_SIZE);
924 _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
927 _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
928 _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
929 _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
932 * Add info to guest to indentify which host is it being run on
933 * and what is the uuid of the guest
935 if (kvmppc_get_host_model(&buf)) {
936 _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
939 if (kvmppc_get_host_serial(&buf)) {
940 _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
944 buf = qemu_uuid_unparse_strdup(&qemu_uuid);
946 _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
948 _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
952 if (qemu_get_vm_name()) {
953 _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
954 qemu_get_vm_name()));
957 _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
958 _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
960 /* /interrupt controller */
961 spapr_dt_xics(spapr->nr_servers, fdt, PHANDLE_XICP);
963 ret = spapr_populate_memory(spapr, fdt);
965 error_report("couldn't setup memory nodes in fdt");
970 spapr_dt_vdevice(spapr->vio_bus, fdt);
972 if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
973 ret = spapr_rng_populate_dt(fdt);
975 error_report("could not set up rng device in the fdt");
980 QLIST_FOREACH(phb, &spapr->phbs, list) {
981 ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
983 error_report("couldn't setup PCI devices in fdt");
989 spapr_populate_cpus_dt_node(fdt, spapr);
991 if (smc->dr_lmb_enabled) {
992 _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
995 if (mc->has_hotpluggable_cpus) {
996 int offset = fdt_path_offset(fdt, "/cpus");
997 ret = spapr_drc_populate_dt(fdt, offset, NULL,
998 SPAPR_DR_CONNECTOR_TYPE_CPU);
1000 error_report("Couldn't set up CPU DR device tree properties");
1005 /* /event-sources */
1006 spapr_dt_events(spapr, fdt);
1009 spapr_dt_rtas(spapr, fdt);
1012 spapr_dt_chosen(spapr, fdt);
1015 if (kvm_enabled()) {
1016 spapr_dt_hypervisor(spapr, fdt);
1019 /* Build memory reserve map */
1020 if (spapr->kernel_size) {
1021 _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1023 if (spapr->initrd_size) {
1024 _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1027 /* ibm,client-architecture-support updates */
1028 ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1030 error_report("couldn't setup CAS properties fdt");
1037 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1039 return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1042 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1045 CPUPPCState *env = &cpu->env;
1047 /* The TCG path should also be holding the BQL at this point */
1048 g_assert(qemu_mutex_iothread_locked());
1051 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1052 env->gpr[3] = H_PRIVILEGE;
1054 env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1058 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1059 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1060 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1061 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1062 #define DIRTY_HPTE(_hpte) ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1065 * Get the fd to access the kernel htab, re-opening it if necessary
1067 static int get_htab_fd(sPAPRMachineState *spapr)
1069 if (spapr->htab_fd >= 0) {
1070 return spapr->htab_fd;
1073 spapr->htab_fd = kvmppc_get_htab_fd(false);
1074 if (spapr->htab_fd < 0) {
1075 error_report("Unable to open fd for reading hash table from KVM: %s",
1079 return spapr->htab_fd;
1082 static void close_htab_fd(sPAPRMachineState *spapr)
1084 if (spapr->htab_fd >= 0) {
1085 close(spapr->htab_fd);
1087 spapr->htab_fd = -1;
1090 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1092 sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1094 return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1097 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1100 sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1101 hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1105 * HTAB is controlled by KVM. Fetch into temporary buffer
1107 ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1108 kvmppc_read_hptes(hptes, ptex, n);
1113 * HTAB is controlled by QEMU. Just point to the internally
1116 return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1119 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1120 const ppc_hash_pte64_t *hptes,
1123 sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1126 g_free((void *)hptes);
1129 /* Nothing to do for qemu managed HPT */
1132 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1133 uint64_t pte0, uint64_t pte1)
1135 sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1136 hwaddr offset = ptex * HASH_PTE_SIZE_64;
1139 kvmppc_write_hpte(ptex, pte0, pte1);
1141 stq_p(spapr->htab + offset, pte0);
1142 stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1146 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1150 /* We aim for a hash table of size 1/128 the size of RAM (rounded
1151 * up). The PAPR recommendation is actually 1/64 of RAM size, but
1152 * that's much more than is needed for Linux guests */
1153 shift = ctz64(pow2ceil(ramsize)) - 7;
1154 shift = MAX(shift, 18); /* Minimum architected size */
1155 shift = MIN(shift, 46); /* Maximum architected size */
1159 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1164 /* Clean up any HPT info from a previous boot */
1165 g_free(spapr->htab);
1167 spapr->htab_shift = 0;
1168 close_htab_fd(spapr);
1170 rc = kvmppc_reset_htab(shift);
1172 /* kernel-side HPT needed, but couldn't allocate one */
1173 error_setg_errno(errp, errno,
1174 "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1176 /* This is almost certainly fatal, but if the caller really
1177 * wants to carry on with shift == 0, it's welcome to try */
1178 } else if (rc > 0) {
1179 /* kernel-side HPT allocated */
1182 "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1186 spapr->htab_shift = shift;
1189 /* kernel-side HPT not needed, allocate in userspace instead */
1190 size_t size = 1ULL << shift;
1193 spapr->htab = qemu_memalign(size, size);
1195 error_setg_errno(errp, errno,
1196 "Could not allocate HPT of order %d", shift);
1200 memset(spapr->htab, 0, size);
1201 spapr->htab_shift = shift;
1203 for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1204 DIRTY_HPTE(HPTE(spapr->htab, i));
1209 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1211 bool matched = false;
1213 if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1218 error_report("Device %s is not supported by this machine yet.",
1219 qdev_fw_name(DEVICE(sbdev)));
1224 static void ppc_spapr_reset(void)
1226 MachineState *machine = MACHINE(qdev_get_machine());
1227 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1228 PowerPCCPU *first_ppc_cpu;
1229 uint32_t rtas_limit;
1230 hwaddr rtas_addr, fdt_addr;
1234 /* Check for unknown sysbus devices */
1235 foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1237 /* Allocate and/or reset the hash page table */
1238 spapr_reallocate_hpt(spapr,
1239 spapr_hpt_shift_for_ramsize(machine->maxram_size),
1242 /* Update the RMA size if necessary */
1243 if (spapr->vrma_adjust) {
1244 spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1248 qemu_devices_reset();
1251 * We place the device tree and RTAS just below either the top of the RMA,
1252 * or just below 2GB, whichever is lowere, so that it can be
1253 * processed with 32-bit real mode code if necessary
1255 rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1256 rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1257 fdt_addr = rtas_addr - FDT_MAX_SIZE;
1259 /* if this reset wasn't generated by CAS, we should reset our
1260 * negotiated options and start from scratch */
1261 if (!spapr->cas_reboot) {
1262 spapr_ovec_cleanup(spapr->ov5_cas);
1263 spapr->ov5_cas = spapr_ovec_new();
1266 fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1268 spapr_load_rtas(spapr, fdt, rtas_addr);
1272 /* Should only fail if we've built a corrupted tree */
1275 if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1276 error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1277 fdt_totalsize(fdt), FDT_MAX_SIZE);
1282 qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1283 cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1286 /* Set up the entry state */
1287 first_ppc_cpu = POWERPC_CPU(first_cpu);
1288 first_ppc_cpu->env.gpr[3] = fdt_addr;
1289 first_ppc_cpu->env.gpr[5] = 0;
1290 first_cpu->halted = 0;
1291 first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1293 spapr->cas_reboot = false;
1296 static void spapr_create_nvram(sPAPRMachineState *spapr)
1298 DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1299 DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1302 qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1306 qdev_init_nofail(dev);
1308 spapr->nvram = (struct sPAPRNVRAM *)dev;
1311 static void spapr_rtc_create(sPAPRMachineState *spapr)
1313 DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1315 qdev_init_nofail(dev);
1318 object_property_add_alias(qdev_get_machine(), "rtc-time",
1319 OBJECT(spapr->rtc), "date", NULL);
1322 /* Returns whether we want to use VGA or not */
1323 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1325 switch (vga_interface_type) {
1332 return pci_vga_init(pci_bus) != NULL;
1335 "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1340 static int spapr_post_load(void *opaque, int version_id)
1342 sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1345 if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1347 for (i = 0; i < spapr->nr_servers; i++) {
1348 icp_resend(&spapr->icps[i]);
1352 /* In earlier versions, there was no separate qdev for the PAPR
1353 * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1354 * So when migrating from those versions, poke the incoming offset
1355 * value into the RTC device */
1356 if (version_id < 3) {
1357 err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1363 static bool version_before_3(void *opaque, int version_id)
1365 return version_id < 3;
1368 static bool spapr_ov5_cas_needed(void *opaque)
1370 sPAPRMachineState *spapr = opaque;
1371 sPAPROptionVector *ov5_mask = spapr_ovec_new();
1372 sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1373 sPAPROptionVector *ov5_removed = spapr_ovec_new();
1376 /* Prior to the introduction of sPAPROptionVector, we had two option
1377 * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1378 * Both of these options encode machine topology into the device-tree
1379 * in such a way that the now-booted OS should still be able to interact
1380 * appropriately with QEMU regardless of what options were actually
1381 * negotiatied on the source side.
1383 * As such, we can avoid migrating the CAS-negotiated options if these
1384 * are the only options available on the current machine/platform.
1385 * Since these are the only options available for pseries-2.7 and
1386 * earlier, this allows us to maintain old->new/new->old migration
1389 * For QEMU 2.8+, there are additional CAS-negotiatable options available
1390 * via default pseries-2.8 machines and explicit command-line parameters.
1391 * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1392 * of the actual CAS-negotiated values to continue working properly. For
1393 * example, availability of memory unplug depends on knowing whether
1394 * OV5_HP_EVT was negotiated via CAS.
1396 * Thus, for any cases where the set of available CAS-negotiatable
1397 * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1398 * include the CAS-negotiated options in the migration stream.
1400 spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1401 spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1403 /* spapr_ovec_diff returns true if bits were removed. we avoid using
1404 * the mask itself since in the future it's possible "legacy" bits may be
1405 * removed via machine options, which could generate a false positive
1406 * that breaks migration.
1408 spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1409 cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1411 spapr_ovec_cleanup(ov5_mask);
1412 spapr_ovec_cleanup(ov5_legacy);
1413 spapr_ovec_cleanup(ov5_removed);
1418 static const VMStateDescription vmstate_spapr_ov5_cas = {
1419 .name = "spapr_option_vector_ov5_cas",
1421 .minimum_version_id = 1,
1422 .needed = spapr_ov5_cas_needed,
1423 .fields = (VMStateField[]) {
1424 VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1425 vmstate_spapr_ovec, sPAPROptionVector),
1426 VMSTATE_END_OF_LIST()
1430 static const VMStateDescription vmstate_spapr = {
1433 .minimum_version_id = 1,
1434 .post_load = spapr_post_load,
1435 .fields = (VMStateField[]) {
1436 /* used to be @next_irq */
1437 VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1440 VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1442 VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1443 VMSTATE_END_OF_LIST()
1445 .subsections = (const VMStateDescription*[]) {
1446 &vmstate_spapr_ov5_cas,
1451 static int htab_save_setup(QEMUFile *f, void *opaque)
1453 sPAPRMachineState *spapr = opaque;
1455 /* "Iteration" header */
1456 qemu_put_be32(f, spapr->htab_shift);
1459 spapr->htab_save_index = 0;
1460 spapr->htab_first_pass = true;
1462 assert(kvm_enabled());
1469 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1472 bool has_timeout = max_ns != -1;
1473 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1474 int index = spapr->htab_save_index;
1475 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1477 assert(spapr->htab_first_pass);
1482 /* Consume invalid HPTEs */
1483 while ((index < htabslots)
1484 && !HPTE_VALID(HPTE(spapr->htab, index))) {
1486 CLEAN_HPTE(HPTE(spapr->htab, index));
1489 /* Consume valid HPTEs */
1491 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1492 && HPTE_VALID(HPTE(spapr->htab, index))) {
1494 CLEAN_HPTE(HPTE(spapr->htab, index));
1497 if (index > chunkstart) {
1498 int n_valid = index - chunkstart;
1500 qemu_put_be32(f, chunkstart);
1501 qemu_put_be16(f, n_valid);
1502 qemu_put_be16(f, 0);
1503 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1504 HASH_PTE_SIZE_64 * n_valid);
1507 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1511 } while ((index < htabslots) && !qemu_file_rate_limit(f));
1513 if (index >= htabslots) {
1514 assert(index == htabslots);
1516 spapr->htab_first_pass = false;
1518 spapr->htab_save_index = index;
1521 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1524 bool final = max_ns < 0;
1525 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1526 int examined = 0, sent = 0;
1527 int index = spapr->htab_save_index;
1528 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1530 assert(!spapr->htab_first_pass);
1533 int chunkstart, invalidstart;
1535 /* Consume non-dirty HPTEs */
1536 while ((index < htabslots)
1537 && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1543 /* Consume valid dirty HPTEs */
1544 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1545 && HPTE_DIRTY(HPTE(spapr->htab, index))
1546 && HPTE_VALID(HPTE(spapr->htab, index))) {
1547 CLEAN_HPTE(HPTE(spapr->htab, index));
1552 invalidstart = index;
1553 /* Consume invalid dirty HPTEs */
1554 while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1555 && HPTE_DIRTY(HPTE(spapr->htab, index))
1556 && !HPTE_VALID(HPTE(spapr->htab, index))) {
1557 CLEAN_HPTE(HPTE(spapr->htab, index));
1562 if (index > chunkstart) {
1563 int n_valid = invalidstart - chunkstart;
1564 int n_invalid = index - invalidstart;
1566 qemu_put_be32(f, chunkstart);
1567 qemu_put_be16(f, n_valid);
1568 qemu_put_be16(f, n_invalid);
1569 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1570 HASH_PTE_SIZE_64 * n_valid);
1571 sent += index - chunkstart;
1573 if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1578 if (examined >= htabslots) {
1582 if (index >= htabslots) {
1583 assert(index == htabslots);
1586 } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1588 if (index >= htabslots) {
1589 assert(index == htabslots);
1593 spapr->htab_save_index = index;
1595 return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1598 #define MAX_ITERATION_NS 5000000 /* 5 ms */
1599 #define MAX_KVM_BUF_SIZE 2048
1601 static int htab_save_iterate(QEMUFile *f, void *opaque)
1603 sPAPRMachineState *spapr = opaque;
1607 /* Iteration header */
1608 qemu_put_be32(f, 0);
1611 assert(kvm_enabled());
1613 fd = get_htab_fd(spapr);
1618 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1622 } else if (spapr->htab_first_pass) {
1623 htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1625 rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1629 qemu_put_be32(f, 0);
1630 qemu_put_be16(f, 0);
1631 qemu_put_be16(f, 0);
1636 static int htab_save_complete(QEMUFile *f, void *opaque)
1638 sPAPRMachineState *spapr = opaque;
1641 /* Iteration header */
1642 qemu_put_be32(f, 0);
1647 assert(kvm_enabled());
1649 fd = get_htab_fd(spapr);
1654 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1659 if (spapr->htab_first_pass) {
1660 htab_save_first_pass(f, spapr, -1);
1662 htab_save_later_pass(f, spapr, -1);
1666 qemu_put_be32(f, 0);
1667 qemu_put_be16(f, 0);
1668 qemu_put_be16(f, 0);
1673 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1675 sPAPRMachineState *spapr = opaque;
1676 uint32_t section_hdr;
1679 if (version_id < 1 || version_id > 1) {
1680 error_report("htab_load() bad version");
1684 section_hdr = qemu_get_be32(f);
1687 Error *local_err = NULL;
1689 /* First section gives the htab size */
1690 spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1692 error_report_err(local_err);
1699 assert(kvm_enabled());
1701 fd = kvmppc_get_htab_fd(true);
1703 error_report("Unable to open fd to restore KVM hash table: %s",
1710 uint16_t n_valid, n_invalid;
1712 index = qemu_get_be32(f);
1713 n_valid = qemu_get_be16(f);
1714 n_invalid = qemu_get_be16(f);
1716 if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1721 if ((index + n_valid + n_invalid) >
1722 (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1723 /* Bad index in stream */
1725 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1726 index, n_valid, n_invalid, spapr->htab_shift);
1732 qemu_get_buffer(f, HPTE(spapr->htab, index),
1733 HASH_PTE_SIZE_64 * n_valid);
1736 memset(HPTE(spapr->htab, index + n_valid), 0,
1737 HASH_PTE_SIZE_64 * n_invalid);
1744 rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1759 static void htab_cleanup(void *opaque)
1761 sPAPRMachineState *spapr = opaque;
1763 close_htab_fd(spapr);
1766 static SaveVMHandlers savevm_htab_handlers = {
1767 .save_live_setup = htab_save_setup,
1768 .save_live_iterate = htab_save_iterate,
1769 .save_live_complete_precopy = htab_save_complete,
1770 .cleanup = htab_cleanup,
1771 .load_state = htab_load,
1774 static void spapr_boot_set(void *opaque, const char *boot_device,
1777 MachineState *machine = MACHINE(qdev_get_machine());
1778 machine->boot_order = g_strdup(boot_device);
1782 * Reset routine for LMB DR devices.
1784 * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1785 * routine. Reset for PCI DR devices will be handled by PHB reset routine
1786 * when it walks all its children devices. LMB devices reset occurs
1787 * as part of spapr_ppc_reset().
1789 static void spapr_drc_reset(void *opaque)
1791 sPAPRDRConnector *drc = opaque;
1792 DeviceState *d = DEVICE(drc);
1799 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1801 MachineState *machine = MACHINE(spapr);
1802 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1803 uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1806 for (i = 0; i < nr_lmbs; i++) {
1807 sPAPRDRConnector *drc;
1810 addr = i * lmb_size + spapr->hotplug_memory.base;
1811 drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1813 qemu_register_reset(spapr_drc_reset, drc);
1818 * If RAM size, maxmem size and individual node mem sizes aren't aligned
1819 * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1820 * since we can't support such unaligned sizes with DRCONF_MEMORY.
1822 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1826 if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1827 error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1828 " is not aligned to %llu MiB",
1830 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1834 if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1835 error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1836 " is not aligned to %llu MiB",
1838 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1842 for (i = 0; i < nb_numa_nodes; i++) {
1843 if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1845 "Node %d memory size 0x%" PRIx64
1846 " is not aligned to %llu MiB",
1847 i, numa_info[i].node_mem,
1848 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1854 /* find cpu slot in machine->possible_cpus by core_id */
1855 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
1857 int index = id / smp_threads;
1859 if (index >= ms->possible_cpus->len) {
1865 return &ms->possible_cpus->cpus[index];
1868 static void spapr_init_cpus(sPAPRMachineState *spapr)
1870 MachineState *machine = MACHINE(spapr);
1871 MachineClass *mc = MACHINE_GET_CLASS(machine);
1872 char *type = spapr_get_cpu_core_type(machine->cpu_model);
1873 int smt = kvmppc_smt_threads();
1874 const CPUArchIdList *possible_cpus;
1875 int boot_cores_nr = smp_cpus / smp_threads;
1879 error_report("Unable to find sPAPR CPU Core definition");
1883 possible_cpus = mc->possible_cpu_arch_ids(machine);
1884 if (mc->has_hotpluggable_cpus) {
1885 if (smp_cpus % smp_threads) {
1886 error_report("smp_cpus (%u) must be multiple of threads (%u)",
1887 smp_cpus, smp_threads);
1890 if (max_cpus % smp_threads) {
1891 error_report("max_cpus (%u) must be multiple of threads (%u)",
1892 max_cpus, smp_threads);
1896 if (max_cpus != smp_cpus) {
1897 error_report("This machine version does not support CPU hotplug");
1900 boot_cores_nr = possible_cpus->len;
1903 for (i = 0; i < possible_cpus->len; i++) {
1904 int core_id = i * smp_threads;
1906 if (mc->has_hotpluggable_cpus) {
1907 sPAPRDRConnector *drc =
1908 spapr_dr_connector_new(OBJECT(spapr),
1909 SPAPR_DR_CONNECTOR_TYPE_CPU,
1910 (core_id / smp_threads) * smt);
1912 qemu_register_reset(spapr_drc_reset, drc);
1915 if (i < boot_cores_nr) {
1916 Object *core = object_new(type);
1917 int nr_threads = smp_threads;
1919 /* Handle the partially filled core for older machine types */
1920 if ((i + 1) * smp_threads >= smp_cpus) {
1921 nr_threads = smp_cpus - i * smp_threads;
1924 object_property_set_int(core, nr_threads, "nr-threads",
1926 object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
1928 object_property_set_bool(core, true, "realized", &error_fatal);
1934 /* pSeries LPAR / sPAPR hardware init */
1935 static void ppc_spapr_init(MachineState *machine)
1937 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1938 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1939 const char *kernel_filename = machine->kernel_filename;
1940 const char *initrd_filename = machine->initrd_filename;
1943 MemoryRegion *sysmem = get_system_memory();
1944 MemoryRegion *ram = g_new(MemoryRegion, 1);
1945 MemoryRegion *rma_region;
1947 hwaddr rma_alloc_size;
1948 hwaddr node0_size = spapr_node0_size();
1949 long load_limit, fw_size;
1951 int smt = kvmppc_smt_threads();
1953 msi_nonbroken = true;
1955 QLIST_INIT(&spapr->phbs);
1957 /* Allocate RMA if necessary */
1958 rma_alloc_size = kvmppc_alloc_rma(&rma);
1960 if (rma_alloc_size == -1) {
1961 error_report("Unable to create RMA");
1965 if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1966 spapr->rma_size = rma_alloc_size;
1968 spapr->rma_size = node0_size;
1970 /* With KVM, we don't actually know whether KVM supports an
1971 * unbounded RMA (PR KVM) or is limited by the hash table size
1972 * (HV KVM using VRMA), so we always assume the latter
1974 * In that case, we also limit the initial allocations for RTAS
1975 * etc... to 256M since we have no way to know what the VRMA size
1976 * is going to be as it depends on the size of the hash table
1977 * isn't determined yet.
1979 if (kvm_enabled()) {
1980 spapr->vrma_adjust = 1;
1981 spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1984 /* Actually we don't support unbounded RMA anymore since we
1985 * added proper emulation of HV mode. The max we can get is
1986 * 16G which also happens to be what we configure for PAPR
1987 * mode so make sure we don't do anything bigger than that
1989 spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
1992 if (spapr->rma_size > node0_size) {
1993 error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
1998 /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1999 load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2001 /* Set up Interrupt Controller before we create the VCPUs */
2002 xics_system_init(machine, DIV_ROUND_UP(max_cpus * smt, smp_threads),
2003 XICS_IRQS_SPAPR, &error_fatal);
2005 /* Set up containers for ibm,client-set-architecture negotiated options */
2006 spapr->ov5 = spapr_ovec_new();
2007 spapr->ov5_cas = spapr_ovec_new();
2009 if (smc->dr_lmb_enabled) {
2010 spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2011 spapr_validate_node_memory(machine, &error_fatal);
2014 spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2016 /* advertise support for dedicated HP event source to guests */
2017 if (spapr->use_hotplug_event_source) {
2018 spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2022 if (machine->cpu_model == NULL) {
2023 machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
2026 ppc_cpu_parse_features(machine->cpu_model);
2028 spapr_init_cpus(spapr);
2030 if (kvm_enabled()) {
2031 /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2032 kvmppc_enable_logical_ci_hcalls();
2033 kvmppc_enable_set_mode_hcall();
2035 /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2036 kvmppc_enable_clear_ref_mod_hcalls();
2040 memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2042 memory_region_add_subregion(sysmem, 0, ram);
2044 if (rma_alloc_size && rma) {
2045 rma_region = g_new(MemoryRegion, 1);
2046 memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2047 rma_alloc_size, rma);
2048 vmstate_register_ram_global(rma_region);
2049 memory_region_add_subregion(sysmem, 0, rma_region);
2052 /* initialize hotplug memory address space */
2053 if (machine->ram_size < machine->maxram_size) {
2054 ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2056 * Limit the number of hotpluggable memory slots to half the number
2057 * slots that KVM supports, leaving the other half for PCI and other
2058 * devices. However ensure that number of slots doesn't drop below 32.
2060 int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2061 SPAPR_MAX_RAM_SLOTS;
2063 if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2064 max_memslots = SPAPR_MAX_RAM_SLOTS;
2066 if (machine->ram_slots > max_memslots) {
2067 error_report("Specified number of memory slots %"
2068 PRIu64" exceeds max supported %d",
2069 machine->ram_slots, max_memslots);
2073 spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2074 SPAPR_HOTPLUG_MEM_ALIGN);
2075 memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2076 "hotplug-memory", hotplug_mem_size);
2077 memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2078 &spapr->hotplug_memory.mr);
2081 if (smc->dr_lmb_enabled) {
2082 spapr_create_lmb_dr_connectors(spapr);
2085 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2087 error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2090 spapr->rtas_size = get_image_size(filename);
2091 if (spapr->rtas_size < 0) {
2092 error_report("Could not get size of LPAR rtas '%s'", filename);
2095 spapr->rtas_blob = g_malloc(spapr->rtas_size);
2096 if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2097 error_report("Could not load LPAR rtas '%s'", filename);
2100 if (spapr->rtas_size > RTAS_MAX_SIZE) {
2101 error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2102 (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2107 /* Set up RTAS event infrastructure */
2108 spapr_events_init(spapr);
2110 /* Set up the RTC RTAS interfaces */
2111 spapr_rtc_create(spapr);
2113 /* Set up VIO bus */
2114 spapr->vio_bus = spapr_vio_bus_init();
2116 for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2117 if (serial_hds[i]) {
2118 spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2122 /* We always have at least the nvram device on VIO */
2123 spapr_create_nvram(spapr);
2126 spapr_pci_rtas_init();
2128 phb = spapr_create_phb(spapr, 0);
2130 for (i = 0; i < nb_nics; i++) {
2131 NICInfo *nd = &nd_table[i];
2134 nd->model = g_strdup("ibmveth");
2137 if (strcmp(nd->model, "ibmveth") == 0) {
2138 spapr_vlan_create(spapr->vio_bus, nd);
2140 pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2144 for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2145 spapr_vscsi_create(spapr->vio_bus);
2149 if (spapr_vga_init(phb->bus, &error_fatal)) {
2150 spapr->has_graphics = true;
2151 machine->usb |= defaults_enabled() && !machine->usb_disabled;
2155 if (smc->use_ohci_by_default) {
2156 pci_create_simple(phb->bus, -1, "pci-ohci");
2158 pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2161 if (spapr->has_graphics) {
2162 USBBus *usb_bus = usb_bus_find(-1);
2164 usb_create_simple(usb_bus, "usb-kbd");
2165 usb_create_simple(usb_bus, "usb-mouse");
2169 if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2171 "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2176 if (kernel_filename) {
2177 uint64_t lowaddr = 0;
2179 spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2180 NULL, NULL, &lowaddr, NULL, 1,
2181 PPC_ELF_MACHINE, 0, 0);
2182 if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2183 spapr->kernel_size = load_elf(kernel_filename,
2184 translate_kernel_address, NULL, NULL,
2185 &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2187 spapr->kernel_le = spapr->kernel_size > 0;
2189 if (spapr->kernel_size < 0) {
2190 error_report("error loading %s: %s", kernel_filename,
2191 load_elf_strerror(spapr->kernel_size));
2196 if (initrd_filename) {
2197 /* Try to locate the initrd in the gap between the kernel
2198 * and the firmware. Add a bit of space just in case
2200 spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2201 + 0x1ffff) & ~0xffff;
2202 spapr->initrd_size = load_image_targphys(initrd_filename,
2205 - spapr->initrd_base);
2206 if (spapr->initrd_size < 0) {
2207 error_report("could not load initial ram disk '%s'",
2214 if (bios_name == NULL) {
2215 bios_name = FW_FILE_NAME;
2217 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2219 error_report("Could not find LPAR firmware '%s'", bios_name);
2222 fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2224 error_report("Could not load LPAR firmware '%s'", filename);
2229 /* FIXME: Should register things through the MachineState's qdev
2230 * interface, this is a legacy from the sPAPREnvironment structure
2231 * which predated MachineState but had a similar function */
2232 vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2233 register_savevm_live(NULL, "spapr/htab", -1, 1,
2234 &savevm_htab_handlers, spapr);
2237 QTAILQ_INIT(&spapr->ccs_list);
2238 qemu_register_reset(spapr_ccs_reset_hook, spapr);
2240 qemu_register_boot_set(spapr_boot_set, spapr);
2242 /* to stop and start vmclock */
2243 if (kvm_enabled()) {
2244 qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2249 static int spapr_kvm_type(const char *vm_type)
2255 if (!strcmp(vm_type, "HV")) {
2259 if (!strcmp(vm_type, "PR")) {
2263 error_report("Unknown kvm-type specified '%s'", vm_type);
2268 * Implementation of an interface to adjust firmware path
2269 * for the bootindex property handling.
2271 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2274 #define CAST(type, obj, name) \
2275 ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2276 SCSIDevice *d = CAST(SCSIDevice, dev, TYPE_SCSI_DEVICE);
2277 sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2280 void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2281 VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2282 USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2286 * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2287 * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2288 * in the top 16 bits of the 64-bit LUN
2290 unsigned id = 0x8000 | (d->id << 8) | d->lun;
2291 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2292 (uint64_t)id << 48);
2293 } else if (virtio) {
2295 * We use SRP luns of the form 01000000 | (target << 8) | lun
2296 * in the top 32 bits of the 64-bit LUN
2297 * Note: the quote above is from SLOF and it is wrong,
2298 * the actual binding is:
2299 * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2301 unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2302 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2303 (uint64_t)id << 32);
2306 * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2307 * in the top 32 bits of the 64-bit LUN
2309 unsigned usb_port = atoi(usb->port->path);
2310 unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2311 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2312 (uint64_t)id << 32);
2317 * SLOF probes the USB devices, and if it recognizes that the device is a
2318 * storage device, it changes its name to "storage" instead of "usb-host",
2319 * and additionally adds a child node for the SCSI LUN, so the correct
2320 * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2322 if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2323 USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2324 if (usb_host_dev_is_scsi_storage(usbdev)) {
2325 return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2330 /* Replace "pci" with "pci@800000020000000" */
2331 return g_strdup_printf("pci@%"PRIX64, phb->buid);
2337 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2339 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2341 return g_strdup(spapr->kvm_type);
2344 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2346 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2348 g_free(spapr->kvm_type);
2349 spapr->kvm_type = g_strdup(value);
2352 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2354 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2356 return spapr->use_hotplug_event_source;
2359 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2362 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2364 spapr->use_hotplug_event_source = value;
2367 static void spapr_machine_initfn(Object *obj)
2369 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2371 spapr->htab_fd = -1;
2372 spapr->use_hotplug_event_source = true;
2373 object_property_add_str(obj, "kvm-type",
2374 spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2375 object_property_set_description(obj, "kvm-type",
2376 "Specifies the KVM virtualization mode (HV, PR)",
2378 object_property_add_bool(obj, "modern-hotplug-events",
2379 spapr_get_modern_hotplug_events,
2380 spapr_set_modern_hotplug_events,
2382 object_property_set_description(obj, "modern-hotplug-events",
2383 "Use dedicated hotplug event mechanism in"
2384 " place of standard EPOW events when possible"
2385 " (required for memory hot-unplug support)",
2389 static void spapr_machine_finalizefn(Object *obj)
2391 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2393 g_free(spapr->kvm_type);
2396 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2398 cpu_synchronize_state(cs);
2399 ppc_cpu_do_system_reset(cs);
2402 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2407 async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2411 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2412 uint32_t node, bool dedicated_hp_event_source,
2415 sPAPRDRConnector *drc;
2416 sPAPRDRConnectorClass *drck;
2417 uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2418 int i, fdt_offset, fdt_size;
2420 uint64_t addr = addr_start;
2422 for (i = 0; i < nr_lmbs; i++) {
2423 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2424 addr/SPAPR_MEMORY_BLOCK_SIZE);
2427 fdt = create_device_tree(&fdt_size);
2428 fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2429 SPAPR_MEMORY_BLOCK_SIZE);
2431 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2432 drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2433 addr += SPAPR_MEMORY_BLOCK_SIZE;
2434 if (!dev->hotplugged) {
2435 /* guests expect coldplugged LMBs to be pre-allocated */
2436 drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2437 drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2440 /* send hotplug notification to the
2441 * guest only in case of hotplugged memory
2443 if (dev->hotplugged) {
2444 if (dedicated_hp_event_source) {
2445 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2446 addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2447 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2448 spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2450 drck->get_index(drc));
2452 spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2458 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2459 uint32_t node, Error **errp)
2461 Error *local_err = NULL;
2462 sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2463 PCDIMMDevice *dimm = PC_DIMM(dev);
2464 PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2465 MemoryRegion *mr = ddc->get_memory_region(dimm);
2466 uint64_t align = memory_region_get_alignment(mr);
2467 uint64_t size = memory_region_size(mr);
2471 if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2472 error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2473 "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2477 mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2478 if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2479 error_setg(&local_err, "Memory backend has bad page size. "
2480 "Use 'memory-backend-file' with correct mem-path.");
2484 pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2489 addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2491 pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2495 spapr_add_lmbs(dev, addr, size, node,
2496 spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2500 error_propagate(errp, local_err);
2503 typedef struct sPAPRDIMMState {
2507 static void spapr_lmb_release(DeviceState *dev, void *opaque)
2509 sPAPRDIMMState *ds = (sPAPRDIMMState *)opaque;
2510 HotplugHandler *hotplug_ctrl;
2512 if (--ds->nr_lmbs) {
2519 * Now that all the LMBs have been removed by the guest, call the
2520 * pc-dimm unplug handler to cleanup up the pc-dimm device.
2522 hotplug_ctrl = qdev_get_hotplug_handler(dev);
2523 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2526 static void spapr_del_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2529 sPAPRDRConnector *drc;
2530 sPAPRDRConnectorClass *drck;
2531 uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2533 sPAPRDIMMState *ds = g_malloc0(sizeof(sPAPRDIMMState));
2534 uint64_t addr = addr_start;
2536 ds->nr_lmbs = nr_lmbs;
2537 for (i = 0; i < nr_lmbs; i++) {
2538 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2539 addr / SPAPR_MEMORY_BLOCK_SIZE);
2542 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2543 drck->detach(drc, dev, spapr_lmb_release, ds, errp);
2544 addr += SPAPR_MEMORY_BLOCK_SIZE;
2547 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2548 addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2549 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2550 spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2552 drck->get_index(drc));
2555 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2558 sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2559 PCDIMMDevice *dimm = PC_DIMM(dev);
2560 PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2561 MemoryRegion *mr = ddc->get_memory_region(dimm);
2563 pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2564 object_unparent(OBJECT(dev));
2567 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2568 DeviceState *dev, Error **errp)
2570 Error *local_err = NULL;
2571 PCDIMMDevice *dimm = PC_DIMM(dev);
2572 PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2573 MemoryRegion *mr = ddc->get_memory_region(dimm);
2574 uint64_t size = memory_region_size(mr);
2577 addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2582 spapr_del_lmbs(dev, addr, size, &error_abort);
2584 error_propagate(errp, local_err);
2587 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2588 sPAPRMachineState *spapr)
2590 PowerPCCPU *cpu = POWERPC_CPU(cs);
2591 DeviceClass *dc = DEVICE_GET_CLASS(cs);
2592 int id = ppc_get_vcpu_dt_id(cpu);
2594 int offset, fdt_size;
2597 fdt = create_device_tree(&fdt_size);
2598 nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2599 offset = fdt_add_subnode(fdt, 0, nodename);
2601 spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2604 *fdt_offset = offset;
2608 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2611 MachineState *ms = MACHINE(qdev_get_machine());
2612 CPUCore *cc = CPU_CORE(dev);
2613 CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
2615 core_slot->cpu = NULL;
2616 object_unparent(OBJECT(dev));
2619 static void spapr_core_release(DeviceState *dev, void *opaque)
2621 HotplugHandler *hotplug_ctrl;
2623 hotplug_ctrl = qdev_get_hotplug_handler(dev);
2624 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2628 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
2632 sPAPRDRConnector *drc;
2633 sPAPRDRConnectorClass *drck;
2634 Error *local_err = NULL;
2635 CPUCore *cc = CPU_CORE(dev);
2636 int smt = kvmppc_smt_threads();
2638 if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
2639 error_setg(errp, "Unable to find CPU core with core-id: %d",
2644 error_setg(errp, "Boot CPU core may not be unplugged");
2648 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2651 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2652 drck->detach(drc, dev, spapr_core_release, NULL, &local_err);
2654 error_propagate(errp, local_err);
2658 spapr_hotplug_req_remove_by_index(drc);
2661 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2664 sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
2665 MachineClass *mc = MACHINE_GET_CLASS(spapr);
2666 sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
2667 CPUCore *cc = CPU_CORE(dev);
2668 CPUState *cs = CPU(core->threads);
2669 sPAPRDRConnector *drc;
2670 Error *local_err = NULL;
2673 int smt = kvmppc_smt_threads();
2674 CPUArchId *core_slot;
2677 core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2679 error_setg(errp, "Unable to find CPU core with core-id: %d",
2683 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2685 g_assert(drc || !mc->has_hotpluggable_cpus);
2688 * Setup CPU DT entries only for hotplugged CPUs. For boot time or
2689 * coldplugged CPUs DT entries are setup in spapr_build_fdt().
2691 if (dev->hotplugged) {
2692 fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
2696 sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2697 drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, &local_err);
2700 error_propagate(errp, local_err);
2705 if (dev->hotplugged) {
2707 * Send hotplug notification interrupt to the guest only in case
2708 * of hotplugged CPUs.
2710 spapr_hotplug_req_add_by_index(drc);
2713 * Set the right DRC states for cold plugged CPU.
2716 sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2717 drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2718 drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2721 core_slot->cpu = OBJECT(dev);
2724 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2727 MachineState *machine = MACHINE(OBJECT(hotplug_dev));
2728 MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
2729 Error *local_err = NULL;
2730 CPUCore *cc = CPU_CORE(dev);
2731 char *base_core_type = spapr_get_cpu_core_type(machine->cpu_model);
2732 const char *type = object_get_typename(OBJECT(dev));
2733 CPUArchId *core_slot;
2736 if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
2737 error_setg(&local_err, "CPU hotplug not supported for this machine");
2741 if (strcmp(base_core_type, type)) {
2742 error_setg(&local_err, "CPU core type should be %s", base_core_type);
2746 if (cc->core_id % smp_threads) {
2747 error_setg(&local_err, "invalid core id %d", cc->core_id);
2751 core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2753 error_setg(&local_err, "core id %d out of range", cc->core_id);
2757 if (core_slot->cpu) {
2758 error_setg(&local_err, "core %d already populated", cc->core_id);
2763 g_free(base_core_type);
2764 error_propagate(errp, local_err);
2767 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2768 DeviceState *dev, Error **errp)
2770 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2772 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2775 if (!smc->dr_lmb_enabled) {
2776 error_setg(errp, "Memory hotplug not supported for this machine");
2779 node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2783 if (node < 0 || node >= MAX_NODES) {
2784 error_setg(errp, "Invaild node %d", node);
2789 * Currently PowerPC kernel doesn't allow hot-adding memory to
2790 * memory-less node, but instead will silently add the memory
2791 * to the first node that has some memory. This causes two
2792 * unexpected behaviours for the user.
2794 * - Memory gets hotplugged to a different node than what the user
2796 * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2797 * to memory-less node, a reboot will set things accordingly
2798 * and the previously hotplugged memory now ends in the right node.
2799 * This appears as if some memory moved from one node to another.
2801 * So until kernel starts supporting memory hotplug to memory-less
2802 * nodes, just prevent such attempts upfront in QEMU.
2804 if (nb_numa_nodes && !numa_info[node].node_mem) {
2805 error_setg(errp, "Can't hotplug memory to memory-less node %d",
2810 spapr_memory_plug(hotplug_dev, dev, node, errp);
2811 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2812 spapr_core_plug(hotplug_dev, dev, errp);
2816 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2817 DeviceState *dev, Error **errp)
2819 sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2820 MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2822 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2823 if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2824 spapr_memory_unplug(hotplug_dev, dev, errp);
2826 error_setg(errp, "Memory hot unplug not supported for this guest");
2828 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2829 if (!mc->has_hotpluggable_cpus) {
2830 error_setg(errp, "CPU hot unplug not supported on this machine");
2833 spapr_core_unplug(hotplug_dev, dev, errp);
2837 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
2838 DeviceState *dev, Error **errp)
2840 sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2841 MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2843 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2844 if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2845 spapr_memory_unplug_request(hotplug_dev, dev, errp);
2847 /* NOTE: this means there is a window after guest reset, prior to
2848 * CAS negotiation, where unplug requests will fail due to the
2849 * capability not being detected yet. This is a bit different than
2850 * the case with PCI unplug, where the events will be queued and
2851 * eventually handled by the guest after boot
2853 error_setg(errp, "Memory hot unplug not supported for this guest");
2855 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2856 if (!mc->has_hotpluggable_cpus) {
2857 error_setg(errp, "CPU hot unplug not supported on this machine");
2860 spapr_core_unplug_request(hotplug_dev, dev, errp);
2864 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
2865 DeviceState *dev, Error **errp)
2867 if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2868 spapr_core_pre_plug(hotplug_dev, dev, errp);
2872 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
2875 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
2876 object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2877 return HOTPLUG_HANDLER(machine);
2882 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index)
2884 /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2885 * socket means much for the paravirtualized PAPR platform) */
2886 return cpu_index / smp_threads / smp_cores;
2889 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
2892 int spapr_max_cores = max_cpus / smp_threads;
2893 MachineClass *mc = MACHINE_GET_CLASS(machine);
2895 if (!mc->has_hotpluggable_cpus) {
2896 spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
2898 if (machine->possible_cpus) {
2899 assert(machine->possible_cpus->len == spapr_max_cores);
2900 return machine->possible_cpus;
2903 machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
2904 sizeof(CPUArchId) * spapr_max_cores);
2905 machine->possible_cpus->len = spapr_max_cores;
2906 for (i = 0; i < machine->possible_cpus->len; i++) {
2907 int core_id = i * smp_threads;
2909 machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
2910 machine->possible_cpus->cpus[i].arch_id = core_id;
2911 machine->possible_cpus->cpus[i].props.has_core_id = true;
2912 machine->possible_cpus->cpus[i].props.core_id = core_id;
2913 /* TODO: add 'has_node/node' here to describe
2914 to which node core belongs */
2916 return machine->possible_cpus;
2919 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
2920 uint64_t *buid, hwaddr *pio,
2921 hwaddr *mmio32, hwaddr *mmio64,
2922 unsigned n_dma, uint32_t *liobns, Error **errp)
2925 * New-style PHB window placement.
2927 * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
2928 * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
2931 * Some guest kernels can't work with MMIO windows above 1<<46
2932 * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
2934 * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
2935 * PHB stacked together. (32TiB+2GiB)..(32TiB+64GiB) contains the
2936 * 2GiB 32-bit MMIO windows for each PHB. Then 33..64TiB has the
2937 * 1TiB 64-bit MMIO windows for each PHB.
2939 const uint64_t base_buid = 0x800000020000000ULL;
2940 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
2941 SPAPR_PCI_MEM64_WIN_SIZE - 1)
2944 /* Sanity check natural alignments */
2945 QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2946 QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2947 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
2948 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
2949 /* Sanity check bounds */
2950 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
2951 SPAPR_PCI_MEM32_WIN_SIZE);
2952 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
2953 SPAPR_PCI_MEM64_WIN_SIZE);
2955 if (index >= SPAPR_MAX_PHBS) {
2956 error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
2957 SPAPR_MAX_PHBS - 1);
2961 *buid = base_buid + index;
2962 for (i = 0; i < n_dma; ++i) {
2963 liobns[i] = SPAPR_PCI_LIOBN(index, i);
2966 *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
2967 *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
2968 *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
2971 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
2973 sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
2975 return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
2978 static void spapr_ics_resend(XICSFabric *dev)
2980 sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
2982 ics_resend(spapr->ics);
2985 static ICPState *spapr_icp_get(XICSFabric *xi, int server)
2987 sPAPRMachineState *spapr = SPAPR_MACHINE(xi);
2989 return (server < spapr->nr_servers) ? &spapr->icps[server] : NULL;
2992 static void spapr_pic_print_info(InterruptStatsProvider *obj,
2995 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2998 for (i = 0; i < spapr->nr_servers; i++) {
2999 icp_pic_print_info(&spapr->icps[i], mon);
3002 ics_pic_print_info(spapr->ics, mon);
3005 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3007 MachineClass *mc = MACHINE_CLASS(oc);
3008 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3009 FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3010 NMIClass *nc = NMI_CLASS(oc);
3011 HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3012 PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3013 XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3014 InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3016 mc->desc = "pSeries Logical Partition (PAPR compliant)";
3019 * We set up the default / latest behaviour here. The class_init
3020 * functions for the specific versioned machine types can override
3021 * these details for backwards compatibility
3023 mc->init = ppc_spapr_init;
3024 mc->reset = ppc_spapr_reset;
3025 mc->block_default_type = IF_SCSI;
3026 mc->max_cpus = 1024;
3027 mc->no_parallel = 1;
3028 mc->default_boot_order = "";
3029 mc->default_ram_size = 512 * M_BYTE;
3030 mc->kvm_type = spapr_kvm_type;
3031 mc->has_dynamic_sysbus = true;
3032 mc->pci_allow_0_address = true;
3033 mc->get_hotplug_handler = spapr_get_hotplug_handler;
3034 hc->pre_plug = spapr_machine_device_pre_plug;
3035 hc->plug = spapr_machine_device_plug;
3036 hc->unplug = spapr_machine_device_unplug;
3037 mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
3038 mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3039 hc->unplug_request = spapr_machine_device_unplug_request;
3041 smc->dr_lmb_enabled = true;
3042 smc->tcg_default_cpu = "POWER8";
3043 mc->has_hotpluggable_cpus = true;
3044 fwc->get_dev_path = spapr_get_fw_dev_path;
3045 nc->nmi_monitor_handler = spapr_nmi;
3046 smc->phb_placement = spapr_phb_placement;
3047 vhc->hypercall = emulate_spapr_hypercall;
3048 vhc->hpt_mask = spapr_hpt_mask;
3049 vhc->map_hptes = spapr_map_hptes;
3050 vhc->unmap_hptes = spapr_unmap_hptes;
3051 vhc->store_hpte = spapr_store_hpte;
3052 xic->ics_get = spapr_ics_get;
3053 xic->ics_resend = spapr_ics_resend;
3054 xic->icp_get = spapr_icp_get;
3055 ispc->print_info = spapr_pic_print_info;
3058 static const TypeInfo spapr_machine_info = {
3059 .name = TYPE_SPAPR_MACHINE,
3060 .parent = TYPE_MACHINE,
3062 .instance_size = sizeof(sPAPRMachineState),
3063 .instance_init = spapr_machine_initfn,
3064 .instance_finalize = spapr_machine_finalizefn,
3065 .class_size = sizeof(sPAPRMachineClass),
3066 .class_init = spapr_machine_class_init,
3067 .interfaces = (InterfaceInfo[]) {
3068 { TYPE_FW_PATH_PROVIDER },
3070 { TYPE_HOTPLUG_HANDLER },
3071 { TYPE_PPC_VIRTUAL_HYPERVISOR },
3072 { TYPE_XICS_FABRIC },
3073 { TYPE_INTERRUPT_STATS_PROVIDER },
3078 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest) \
3079 static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3082 MachineClass *mc = MACHINE_CLASS(oc); \
3083 spapr_machine_##suffix##_class_options(mc); \
3085 mc->alias = "pseries"; \
3086 mc->is_default = 1; \
3089 static void spapr_machine_##suffix##_instance_init(Object *obj) \
3091 MachineState *machine = MACHINE(obj); \
3092 spapr_machine_##suffix##_instance_options(machine); \
3094 static const TypeInfo spapr_machine_##suffix##_info = { \
3095 .name = MACHINE_TYPE_NAME("pseries-" verstr), \
3096 .parent = TYPE_SPAPR_MACHINE, \
3097 .class_init = spapr_machine_##suffix##_class_init, \
3098 .instance_init = spapr_machine_##suffix##_instance_init, \
3100 static void spapr_machine_register_##suffix(void) \
3102 type_register(&spapr_machine_##suffix##_info); \
3104 type_init(spapr_machine_register_##suffix)
3109 static void spapr_machine_2_9_instance_options(MachineState *machine)
3113 static void spapr_machine_2_9_class_options(MachineClass *mc)
3115 /* Defaults for the latest behaviour inherited from the base class */
3118 DEFINE_SPAPR_MACHINE(2_9, "2.9", true);
3123 #define SPAPR_COMPAT_2_8 \
3126 static void spapr_machine_2_8_instance_options(MachineState *machine)
3128 spapr_machine_2_9_instance_options(machine);
3131 static void spapr_machine_2_8_class_options(MachineClass *mc)
3133 spapr_machine_2_9_class_options(mc);
3134 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3137 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3142 #define SPAPR_COMPAT_2_7 \
3145 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
3146 .property = "mem_win_size", \
3147 .value = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3150 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
3151 .property = "mem64_win_size", \
3155 .driver = TYPE_POWERPC_CPU, \
3156 .property = "pre-2.8-migration", \
3160 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
3161 .property = "pre-2.8-migration", \
3165 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3166 uint64_t *buid, hwaddr *pio,
3167 hwaddr *mmio32, hwaddr *mmio64,
3168 unsigned n_dma, uint32_t *liobns, Error **errp)
3170 /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3171 const uint64_t base_buid = 0x800000020000000ULL;
3172 const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3173 const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3174 const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3175 const uint32_t max_index = 255;
3176 const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3178 uint64_t ram_top = MACHINE(spapr)->ram_size;
3179 hwaddr phb0_base, phb_base;
3182 /* Do we have hotpluggable memory? */
3183 if (MACHINE(spapr)->maxram_size > ram_top) {
3184 /* Can't just use maxram_size, because there may be an
3185 * alignment gap between normal and hotpluggable memory
3187 ram_top = spapr->hotplug_memory.base +
3188 memory_region_size(&spapr->hotplug_memory.mr);
3191 phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3193 if (index > max_index) {
3194 error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3199 *buid = base_buid + index;
3200 for (i = 0; i < n_dma; ++i) {
3201 liobns[i] = SPAPR_PCI_LIOBN(index, i);
3204 phb_base = phb0_base + index * phb_spacing;
3205 *pio = phb_base + pio_offset;
3206 *mmio32 = phb_base + mmio_offset;
3208 * We don't set the 64-bit MMIO window, relying on the PHB's
3209 * fallback behaviour of automatically splitting a large "32-bit"
3210 * window into contiguous 32-bit and 64-bit windows
3214 static void spapr_machine_2_7_instance_options(MachineState *machine)
3216 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3218 spapr_machine_2_8_instance_options(machine);
3219 spapr->use_hotplug_event_source = false;
3222 static void spapr_machine_2_7_class_options(MachineClass *mc)
3224 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3226 spapr_machine_2_8_class_options(mc);
3227 smc->tcg_default_cpu = "POWER7";
3228 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3229 smc->phb_placement = phb_placement_2_7;
3232 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3237 #define SPAPR_COMPAT_2_6 \
3240 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3242 .value = stringify(off),\
3245 static void spapr_machine_2_6_instance_options(MachineState *machine)
3247 spapr_machine_2_7_instance_options(machine);
3250 static void spapr_machine_2_6_class_options(MachineClass *mc)
3252 spapr_machine_2_7_class_options(mc);
3253 mc->has_hotpluggable_cpus = false;
3254 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3257 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3262 #define SPAPR_COMPAT_2_5 \
3265 .driver = "spapr-vlan", \
3266 .property = "use-rx-buffer-pools", \
3270 static void spapr_machine_2_5_instance_options(MachineState *machine)
3272 spapr_machine_2_6_instance_options(machine);
3275 static void spapr_machine_2_5_class_options(MachineClass *mc)
3277 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3279 spapr_machine_2_6_class_options(mc);
3280 smc->use_ohci_by_default = true;
3281 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3284 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3289 #define SPAPR_COMPAT_2_4 \
3292 static void spapr_machine_2_4_instance_options(MachineState *machine)
3294 spapr_machine_2_5_instance_options(machine);
3297 static void spapr_machine_2_4_class_options(MachineClass *mc)
3299 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3301 spapr_machine_2_5_class_options(mc);
3302 smc->dr_lmb_enabled = false;
3303 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3306 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3311 #define SPAPR_COMPAT_2_3 \
3314 .driver = "spapr-pci-host-bridge",\
3315 .property = "dynamic-reconfiguration",\
3319 static void spapr_machine_2_3_instance_options(MachineState *machine)
3321 spapr_machine_2_4_instance_options(machine);
3322 savevm_skip_section_footers();
3323 global_state_set_optional();
3324 savevm_skip_configuration();
3327 static void spapr_machine_2_3_class_options(MachineClass *mc)
3329 spapr_machine_2_4_class_options(mc);
3330 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3332 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3338 #define SPAPR_COMPAT_2_2 \
3341 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3342 .property = "mem_win_size",\
3343 .value = "0x20000000",\
3346 static void spapr_machine_2_2_instance_options(MachineState *machine)
3348 spapr_machine_2_3_instance_options(machine);
3349 machine->suppress_vmdesc = true;
3352 static void spapr_machine_2_2_class_options(MachineClass *mc)
3354 spapr_machine_2_3_class_options(mc);
3355 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3357 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3362 #define SPAPR_COMPAT_2_1 \
3365 static void spapr_machine_2_1_instance_options(MachineState *machine)
3367 spapr_machine_2_2_instance_options(machine);
3370 static void spapr_machine_2_1_class_options(MachineClass *mc)
3372 spapr_machine_2_2_class_options(mc);
3373 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3375 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
3377 static void spapr_machine_register_types(void)
3379 type_register_static(&spapr_machine_info);
3382 type_init(spapr_machine_register_types)