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/datadir.h"
29 #include "qemu/memalign.h"
30 #include "qapi/error.h"
31 #include "qapi/qapi-events-machine.h"
32 #include "qapi/qapi-events-qdev.h"
33 #include "qapi/visitor.h"
34 #include "sysemu/sysemu.h"
35 #include "sysemu/hostmem.h"
36 #include "sysemu/numa.h"
37 #include "sysemu/qtest.h"
38 #include "sysemu/reset.h"
39 #include "sysemu/runstate.h"
41 #include "hw/fw-path-provider.h"
44 #include "sysemu/device_tree.h"
45 #include "sysemu/cpus.h"
46 #include "sysemu/hw_accel.h"
48 #include "migration/misc.h"
49 #include "migration/qemu-file-types.h"
50 #include "migration/global_state.h"
51 #include "migration/register.h"
52 #include "migration/blocker.h"
53 #include "mmu-hash64.h"
54 #include "mmu-book3s-v3.h"
55 #include "cpu-models.h"
56 #include "hw/core/cpu.h"
58 #include "hw/ppc/ppc.h"
59 #include "hw/loader.h"
61 #include "hw/ppc/fdt.h"
62 #include "hw/ppc/spapr.h"
63 #include "hw/ppc/spapr_vio.h"
64 #include "hw/qdev-properties.h"
65 #include "hw/pci-host/spapr.h"
66 #include "hw/pci/msi.h"
68 #include "hw/pci/pci.h"
69 #include "hw/scsi/scsi.h"
70 #include "hw/virtio/virtio-scsi.h"
71 #include "hw/virtio/vhost-scsi-common.h"
73 #include "exec/ram_addr.h"
75 #include "qemu/config-file.h"
76 #include "qemu/error-report.h"
79 #include "hw/intc/intc.h"
81 #include "hw/ppc/spapr_cpu_core.h"
82 #include "hw/mem/memory-device.h"
83 #include "hw/ppc/spapr_tpm_proxy.h"
84 #include "hw/ppc/spapr_nvdimm.h"
85 #include "hw/ppc/spapr_numa.h"
86 #include "hw/ppc/pef.h"
88 #include "monitor/monitor.h"
92 /* SLOF memory layout:
94 * SLOF raw image loaded at 0, copies its romfs right below the flat
95 * device-tree, then position SLOF itself 31M below that
97 * So we set FW_OVERHEAD to 40MB which should account for all of that
100 * We load our kernel at 4M, leaving space for SLOF initial image
102 #define FDT_MAX_ADDR 0x80000000 /* FDT must stay below that */
103 #define FW_MAX_SIZE 0x400000
104 #define FW_FILE_NAME "slof.bin"
105 #define FW_FILE_NAME_VOF "vof.bin"
106 #define FW_OVERHEAD 0x2800000
107 #define KERNEL_LOAD_ADDR FW_MAX_SIZE
109 #define MIN_RMA_SLOF (128 * MiB)
111 #define PHANDLE_INTC 0x00001111
113 /* These two functions implement the VCPU id numbering: one to compute them
114 * all and one to identify thread 0 of a VCORE. Any change to the first one
115 * is likely to have an impact on the second one, so let's keep them close.
117 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
119 MachineState *ms = MACHINE(spapr);
120 unsigned int smp_threads = ms->smp.threads;
124 (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
126 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
130 return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
133 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
135 /* Dummy entries correspond to unused ICPState objects in older QEMUs,
136 * and newer QEMUs don't even have them. In both cases, we don't want
137 * to send anything on the wire.
142 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
143 .name = "icp/server",
145 .minimum_version_id = 1,
146 .needed = pre_2_10_vmstate_dummy_icp_needed,
147 .fields = (VMStateField[]) {
148 VMSTATE_UNUSED(4), /* uint32_t xirr */
149 VMSTATE_UNUSED(1), /* uint8_t pending_priority */
150 VMSTATE_UNUSED(1), /* uint8_t mfrr */
151 VMSTATE_END_OF_LIST()
155 static void pre_2_10_vmstate_register_dummy_icp(int i)
157 vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
158 (void *)(uintptr_t) i);
161 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
163 vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
164 (void *)(uintptr_t) i);
167 int spapr_max_server_number(SpaprMachineState *spapr)
169 MachineState *ms = MACHINE(spapr);
172 return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
175 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
179 uint32_t servers_prop[smt_threads];
180 uint32_t gservers_prop[smt_threads * 2];
181 int index = spapr_get_vcpu_id(cpu);
183 if (cpu->compat_pvr) {
184 ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
190 /* Build interrupt servers and gservers properties */
191 for (i = 0; i < smt_threads; i++) {
192 servers_prop[i] = cpu_to_be32(index + i);
193 /* Hack, direct the group queues back to cpu 0 */
194 gservers_prop[i*2] = cpu_to_be32(index + i);
195 gservers_prop[i*2 + 1] = 0;
197 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
198 servers_prop, sizeof(servers_prop));
202 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
203 gservers_prop, sizeof(gservers_prop));
208 static void spapr_dt_pa_features(SpaprMachineState *spapr,
210 void *fdt, int offset)
212 uint8_t pa_features_206[] = { 6, 0,
213 0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
214 uint8_t pa_features_207[] = { 24, 0,
215 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
216 0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
217 0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
218 0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
219 uint8_t pa_features_300[] = { 66, 0,
220 /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
221 /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
222 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
224 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
226 0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
227 /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
228 0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
229 /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
230 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
231 /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
232 0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
233 /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
234 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
235 /* 42: PM, 44: PC RA, 46: SC vec'd */
236 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
237 /* 48: SIMD, 50: QP BFP, 52: String */
238 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
239 /* 54: DecFP, 56: DecI, 58: SHA */
240 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
241 /* 60: NM atomic, 62: RNG */
242 0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
244 uint8_t *pa_features = NULL;
247 if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
248 pa_features = pa_features_206;
249 pa_size = sizeof(pa_features_206);
251 if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
252 pa_features = pa_features_207;
253 pa_size = sizeof(pa_features_207);
255 if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
256 pa_features = pa_features_300;
257 pa_size = sizeof(pa_features_300);
263 if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
265 * Note: we keep CI large pages off by default because a 64K capable
266 * guest provisioned with large pages might otherwise try to map a qemu
267 * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
268 * even if that qemu runs on a 4k host.
269 * We dd this bit back here if we are confident this is not an issue
271 pa_features[3] |= 0x20;
273 if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
274 pa_features[24] |= 0x80; /* Transactional memory support */
276 if (spapr->cas_pre_isa3_guest && pa_size > 40) {
277 /* Workaround for broken kernels that attempt (guest) radix
278 * mode when they can't handle it, if they see the radix bit set
279 * in pa-features. So hide it from them. */
280 pa_features[40 + 2] &= ~0x80; /* Radix MMU */
283 _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
286 static hwaddr spapr_node0_size(MachineState *machine)
288 if (machine->numa_state->num_nodes) {
290 for (i = 0; i < machine->numa_state->num_nodes; ++i) {
291 if (machine->numa_state->nodes[i].node_mem) {
292 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
297 return machine->ram_size;
300 static void add_str(GString *s, const gchar *s1)
302 g_string_append_len(s, s1, strlen(s1) + 1);
305 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
306 hwaddr start, hwaddr size)
309 uint64_t mem_reg_property[2];
312 mem_reg_property[0] = cpu_to_be64(start);
313 mem_reg_property[1] = cpu_to_be64(size);
315 sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
316 off = fdt_add_subnode(fdt, 0, mem_name);
318 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
319 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
320 sizeof(mem_reg_property))));
321 spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
325 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
327 MemoryDeviceInfoList *info;
329 for (info = list; info; info = info->next) {
330 MemoryDeviceInfo *value = info->value;
332 if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
333 PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
335 if (addr >= pcdimm_info->addr &&
336 addr < (pcdimm_info->addr + pcdimm_info->size)) {
337 return pcdimm_info->node;
345 struct sPAPRDrconfCellV2 {
353 typedef struct DrconfCellQueue {
354 struct sPAPRDrconfCellV2 cell;
355 QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
358 static DrconfCellQueue *
359 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
360 uint32_t drc_index, uint32_t aa_index,
363 DrconfCellQueue *elem;
365 elem = g_malloc0(sizeof(*elem));
366 elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
367 elem->cell.base_addr = cpu_to_be64(base_addr);
368 elem->cell.drc_index = cpu_to_be32(drc_index);
369 elem->cell.aa_index = cpu_to_be32(aa_index);
370 elem->cell.flags = cpu_to_be32(flags);
375 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
376 int offset, MemoryDeviceInfoList *dimms)
378 MachineState *machine = MACHINE(spapr);
379 uint8_t *int_buf, *cur_index;
381 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
382 uint64_t addr, cur_addr, size;
383 uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
384 uint64_t mem_end = machine->device_memory->base +
385 memory_region_size(&machine->device_memory->mr);
386 uint32_t node, buf_len, nr_entries = 0;
388 DrconfCellQueue *elem, *next;
389 MemoryDeviceInfoList *info;
390 QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
391 = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
393 /* Entry to cover RAM and the gap area */
394 elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
395 SPAPR_LMB_FLAGS_RESERVED |
396 SPAPR_LMB_FLAGS_DRC_INVALID);
397 QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
400 cur_addr = machine->device_memory->base;
401 for (info = dimms; info; info = info->next) {
402 PCDIMMDeviceInfo *di = info->value->u.dimm.data;
409 * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
410 * area is marked hotpluggable in the next iteration for the bigger
411 * chunk including the NVDIMM occupied area.
413 if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
416 /* Entry for hot-pluggable area */
417 if (cur_addr < addr) {
418 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
420 elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
421 cur_addr, spapr_drc_index(drc), -1, 0);
422 QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
427 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
429 elem = spapr_get_drconf_cell(size / lmb_size, addr,
430 spapr_drc_index(drc), node,
431 (SPAPR_LMB_FLAGS_ASSIGNED |
432 SPAPR_LMB_FLAGS_HOTREMOVABLE));
433 QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
435 cur_addr = addr + size;
438 /* Entry for remaining hotpluggable area */
439 if (cur_addr < mem_end) {
440 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
442 elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
443 cur_addr, spapr_drc_index(drc), -1, 0);
444 QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
448 buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
449 int_buf = cur_index = g_malloc0(buf_len);
450 *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
451 cur_index += sizeof(nr_entries);
453 QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
454 memcpy(cur_index, &elem->cell, sizeof(elem->cell));
455 cur_index += sizeof(elem->cell);
456 QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
460 ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
468 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
469 int offset, MemoryDeviceInfoList *dimms)
471 MachineState *machine = MACHINE(spapr);
473 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
474 uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
475 uint32_t nr_lmbs = (machine->device_memory->base +
476 memory_region_size(&machine->device_memory->mr)) /
478 uint32_t *int_buf, *cur_index, buf_len;
481 * Allocate enough buffer size to fit in ibm,dynamic-memory
483 buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
484 cur_index = int_buf = g_malloc0(buf_len);
485 int_buf[0] = cpu_to_be32(nr_lmbs);
487 for (i = 0; i < nr_lmbs; i++) {
488 uint64_t addr = i * lmb_size;
489 uint32_t *dynamic_memory = cur_index;
491 if (i >= device_lmb_start) {
494 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
497 dynamic_memory[0] = cpu_to_be32(addr >> 32);
498 dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
499 dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
500 dynamic_memory[3] = cpu_to_be32(0); /* reserved */
501 dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
502 if (memory_region_present(get_system_memory(), addr)) {
503 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
505 dynamic_memory[5] = cpu_to_be32(0);
509 * LMB information for RMA, boot time RAM and gap b/n RAM and
510 * device memory region -- all these are marked as reserved
511 * and as having no valid DRC.
513 dynamic_memory[0] = cpu_to_be32(addr >> 32);
514 dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
515 dynamic_memory[2] = cpu_to_be32(0);
516 dynamic_memory[3] = cpu_to_be32(0); /* reserved */
517 dynamic_memory[4] = cpu_to_be32(-1);
518 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
519 SPAPR_LMB_FLAGS_DRC_INVALID);
522 cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
524 ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
533 * Adds ibm,dynamic-reconfiguration-memory node.
534 * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
535 * of this device tree node.
537 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
540 MachineState *machine = MACHINE(spapr);
542 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
543 uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
544 cpu_to_be32(lmb_size & 0xffffffff)};
545 MemoryDeviceInfoList *dimms = NULL;
548 * Don't create the node if there is no device memory
550 if (machine->ram_size == machine->maxram_size) {
554 offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
556 ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
557 sizeof(prop_lmb_size));
562 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
567 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
572 /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
573 dimms = qmp_memory_device_list();
574 if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
575 ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
577 ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
579 qapi_free_MemoryDeviceInfoList(dimms);
585 ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
590 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
592 MachineState *machine = MACHINE(spapr);
593 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
594 hwaddr mem_start, node_size;
595 int i, nb_nodes = machine->numa_state->num_nodes;
596 NodeInfo *nodes = machine->numa_state->nodes;
598 for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
599 if (!nodes[i].node_mem) {
602 if (mem_start >= machine->ram_size) {
605 node_size = nodes[i].node_mem;
606 if (node_size > machine->ram_size - mem_start) {
607 node_size = machine->ram_size - mem_start;
611 /* spapr_machine_init() checks for rma_size <= node0_size
613 spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
614 mem_start += spapr->rma_size;
615 node_size -= spapr->rma_size;
617 for ( ; node_size; ) {
618 hwaddr sizetmp = pow2floor(node_size);
620 /* mem_start != 0 here */
621 if (ctzl(mem_start) < ctzl(sizetmp)) {
622 sizetmp = 1ULL << ctzl(mem_start);
625 spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
626 node_size -= sizetmp;
627 mem_start += sizetmp;
631 /* Generate ibm,dynamic-reconfiguration-memory node if required */
632 if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
635 g_assert(smc->dr_lmb_enabled);
636 ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
645 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
646 SpaprMachineState *spapr)
648 MachineState *ms = MACHINE(spapr);
649 PowerPCCPU *cpu = POWERPC_CPU(cs);
650 CPUPPCState *env = &cpu->env;
651 PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
652 int index = spapr_get_vcpu_id(cpu);
653 uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
654 0xffffffff, 0xffffffff};
655 uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
656 : SPAPR_TIMEBASE_FREQ;
657 uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
658 uint32_t page_sizes_prop[64];
659 size_t page_sizes_prop_size;
660 unsigned int smp_threads = ms->smp.threads;
661 uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
662 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
663 int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
666 uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
669 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
671 drc_index = spapr_drc_index(drc);
672 _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
675 _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
676 _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
678 _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
679 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
680 env->dcache_line_size)));
681 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
682 env->dcache_line_size)));
683 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
684 env->icache_line_size)));
685 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
686 env->icache_line_size)));
688 if (pcc->l1_dcache_size) {
689 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
690 pcc->l1_dcache_size)));
692 warn_report("Unknown L1 dcache size for cpu");
694 if (pcc->l1_icache_size) {
695 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
696 pcc->l1_icache_size)));
698 warn_report("Unknown L1 icache size for cpu");
701 _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
702 _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
703 _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
704 _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
705 _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
706 _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
708 if (ppc_has_spr(cpu, SPR_PURR)) {
709 _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
711 if (ppc_has_spr(cpu, SPR_PURR)) {
712 _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
715 if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
716 _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
717 segs, sizeof(segs))));
720 /* Advertise VSX (vector extensions) if available
721 * 1 == VMX / Altivec available
724 * Only CPUs for which we create core types in spapr_cpu_core.c
725 * are possible, and all of those have VMX */
726 if (env->insns_flags & PPC_ALTIVEC) {
727 if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
728 _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
730 _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
734 /* Advertise DFP (Decimal Floating Point) if available
735 * 0 / no property == no DFP
736 * 1 == DFP available */
737 if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
738 _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
741 page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
742 sizeof(page_sizes_prop));
743 if (page_sizes_prop_size) {
744 _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
745 page_sizes_prop, page_sizes_prop_size)));
748 spapr_dt_pa_features(spapr, cpu, fdt, offset);
750 _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
751 cs->cpu_index / vcpus_per_socket)));
753 _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
754 pft_size_prop, sizeof(pft_size_prop))));
756 if (ms->numa_state->num_nodes > 1) {
757 _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
760 _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
762 if (pcc->radix_page_info) {
763 for (i = 0; i < pcc->radix_page_info->count; i++) {
764 radix_AP_encodings[i] =
765 cpu_to_be32(pcc->radix_page_info->entries[i]);
767 _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
769 pcc->radix_page_info->count *
770 sizeof(radix_AP_encodings[0]))));
774 * We set this property to let the guest know that it can use the large
775 * decrementer and its width in bits.
777 if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
778 _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
779 pcc->lrg_decr_bits)));
782 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
790 cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
792 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
793 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
796 * We walk the CPUs in reverse order to ensure that CPU DT nodes
797 * created by fdt_add_subnode() end up in the right order in FDT
798 * for the guest kernel the enumerate the CPUs correctly.
800 * The CPU list cannot be traversed in reverse order, so we need
806 rev = g_renew(CPUState *, rev, n_cpus + 1);
810 for (i = n_cpus - 1; i >= 0; i--) {
811 CPUState *cs = rev[i];
812 PowerPCCPU *cpu = POWERPC_CPU(cs);
813 int index = spapr_get_vcpu_id(cpu);
814 DeviceClass *dc = DEVICE_GET_CLASS(cs);
815 g_autofree char *nodename = NULL;
818 if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
822 nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
823 offset = fdt_add_subnode(fdt, cpus_offset, nodename);
825 spapr_dt_cpu(cs, fdt, offset, spapr);
831 static int spapr_dt_rng(void *fdt)
836 node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
840 ret = fdt_setprop_string(fdt, node, "device_type",
841 "ibm,platform-facilities");
842 ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
843 ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
845 node = fdt_add_subnode(fdt, node, "ibm,random-v1");
849 ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
854 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
856 MachineState *ms = MACHINE(spapr);
858 GString *hypertas = g_string_sized_new(256);
859 GString *qemu_hypertas = g_string_sized_new(256);
860 uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
861 memory_region_size(&MACHINE(spapr)->device_memory->mr);
862 uint32_t lrdr_capacity[] = {
863 cpu_to_be32(max_device_addr >> 32),
864 cpu_to_be32(max_device_addr & 0xffffffff),
865 cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
866 cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
867 cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
870 _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
873 add_str(hypertas, "hcall-pft");
874 add_str(hypertas, "hcall-term");
875 add_str(hypertas, "hcall-dabr");
876 add_str(hypertas, "hcall-interrupt");
877 add_str(hypertas, "hcall-tce");
878 add_str(hypertas, "hcall-vio");
879 add_str(hypertas, "hcall-splpar");
880 add_str(hypertas, "hcall-join");
881 add_str(hypertas, "hcall-bulk");
882 add_str(hypertas, "hcall-set-mode");
883 add_str(hypertas, "hcall-sprg0");
884 add_str(hypertas, "hcall-copy");
885 add_str(hypertas, "hcall-debug");
886 add_str(hypertas, "hcall-vphn");
887 if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
888 add_str(hypertas, "hcall-rpt-invalidate");
891 add_str(qemu_hypertas, "hcall-memop1");
893 if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
894 add_str(hypertas, "hcall-multi-tce");
897 if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
898 add_str(hypertas, "hcall-hpt-resize");
901 _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
902 hypertas->str, hypertas->len));
903 g_string_free(hypertas, TRUE);
904 _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
905 qemu_hypertas->str, qemu_hypertas->len));
906 g_string_free(qemu_hypertas, TRUE);
908 spapr_numa_write_rtas_dt(spapr, fdt, rtas);
911 * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
912 * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
914 * The system reset requirements are driven by existing Linux and PowerVM
915 * implementation which (contrary to PAPR) saves r3 in the error log
916 * structure like machine check, so Linux expects to find the saved r3
917 * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
918 * does not look at the error value).
920 * System reset interrupts are not subject to interlock like machine
921 * check, so this memory area could be corrupted if the sreset is
922 * interrupted by a machine check (or vice versa) if it was shared. To
923 * prevent this, system reset uses per-CPU areas for the sreset save
924 * area. A system reset that interrupts a system reset handler could
925 * still overwrite this area, but Linux doesn't try to recover in that
928 * The extra 8 bytes is required because Linux's FWNMI error log check
931 * RTAS_MIN_SIZE is required for the RTAS blob itself.
933 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
935 ms->smp.max_cpus * sizeof(uint64_t) * 2 +
937 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
938 RTAS_ERROR_LOG_MAX));
939 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
940 RTAS_EVENT_SCAN_RATE));
942 g_assert(msi_nonbroken);
943 _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
946 * According to PAPR, rtas ibm,os-term does not guarantee a return
947 * back to the guest cpu.
949 * While an additional ibm,extended-os-term property indicates
950 * that rtas call return will always occur. Set this property.
952 _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
954 _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
955 lrdr_capacity, sizeof(lrdr_capacity)));
957 spapr_dt_rtas_tokens(fdt, rtas);
961 * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
962 * and the XIVE features that the guest may request and thus the valid
963 * values for bytes 23..26 of option vector 5:
965 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
968 PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
971 23, 0x00, /* XICS / XIVE mode */
972 24, 0x00, /* Hash/Radix, filled in below. */
973 25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
974 26, 0x40, /* Radix options: GTSE == yes. */
977 if (spapr->irq->xics && spapr->irq->xive) {
978 val[1] = SPAPR_OV5_XIVE_BOTH;
979 } else if (spapr->irq->xive) {
980 val[1] = SPAPR_OV5_XIVE_EXPLOIT;
982 assert(spapr->irq->xics);
983 val[1] = SPAPR_OV5_XIVE_LEGACY;
986 if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
987 first_ppc_cpu->compat_pvr)) {
989 * If we're in a pre POWER9 compat mode then the guest should
990 * do hash and use the legacy interrupt mode
992 val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
993 val[3] = 0x00; /* Hash */
994 spapr_check_mmu_mode(false);
995 } else if (kvm_enabled()) {
996 if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
997 val[3] = 0x80; /* OV5_MMU_BOTH */
998 } else if (kvmppc_has_cap_mmu_radix()) {
999 val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1001 val[3] = 0x00; /* Hash */
1004 /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1007 _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1011 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1013 MachineState *machine = MACHINE(spapr);
1014 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1017 _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1020 const char *boot_device = spapr->boot_device;
1021 g_autofree char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1023 g_autofree char *bootlist = get_boot_devices_list(&cb);
1025 if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1026 _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1027 machine->kernel_cmdline));
1030 if (spapr->initrd_size) {
1031 _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1032 spapr->initrd_base));
1033 _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1034 spapr->initrd_base + spapr->initrd_size));
1037 if (spapr->kernel_size) {
1038 uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1039 cpu_to_be64(spapr->kernel_size) };
1041 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1042 &kprop, sizeof(kprop)));
1043 if (spapr->kernel_le) {
1044 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1047 if (machine->boot_config.has_menu && machine->boot_config.menu) {
1048 _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", true)));
1050 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1051 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1052 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1054 if (cb && bootlist) {
1057 for (i = 0; i < cb; i++) {
1058 if (bootlist[i] == '\n') {
1062 _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1065 if (boot_device && strlen(boot_device)) {
1066 _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1069 if (spapr->want_stdout_path && stdout_path) {
1071 * "linux,stdout-path" and "stdout" properties are
1072 * deprecated by linux kernel. New platforms should only
1073 * use the "stdout-path" property. Set the new property
1074 * and continue using older property to remain compatible
1075 * with the existing firmware.
1077 _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1078 _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1082 * We can deal with BAR reallocation just fine, advertise it
1085 if (smc->linux_pci_probe) {
1086 _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1089 spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1092 _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1095 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1097 /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1098 * KVM to work under pHyp with some guest co-operation */
1100 uint8_t hypercall[16];
1102 _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1103 /* indicate KVM hypercall interface */
1104 _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1105 if (kvmppc_has_cap_fixup_hcalls()) {
1107 * Older KVM versions with older guest kernels were broken
1108 * with the magic page, don't allow the guest to map it.
1110 if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1111 sizeof(hypercall))) {
1112 _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1113 hypercall, sizeof(hypercall)));
1118 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1120 MachineState *machine = MACHINE(spapr);
1121 MachineClass *mc = MACHINE_GET_CLASS(machine);
1122 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1123 uint32_t root_drc_type_mask = 0;
1129 fdt = g_malloc0(space);
1130 _FDT((fdt_create_empty_tree(fdt, space)));
1133 _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1134 _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1135 _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1137 /* Guest UUID & Name*/
1138 buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1139 _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1140 if (qemu_uuid_set) {
1141 _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1145 if (qemu_get_vm_name()) {
1146 _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1147 qemu_get_vm_name()));
1150 /* Host Model & Serial Number */
1151 if (spapr->host_model) {
1152 _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1153 } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1154 _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1158 if (spapr->host_serial) {
1159 _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1160 } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1161 _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1165 _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1166 _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1168 /* /interrupt controller */
1169 spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1171 ret = spapr_dt_memory(spapr, fdt);
1173 error_report("couldn't setup memory nodes in fdt");
1178 spapr_dt_vdevice(spapr->vio_bus, fdt);
1180 if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1181 ret = spapr_dt_rng(fdt);
1183 error_report("could not set up rng device in the fdt");
1188 QLIST_FOREACH(phb, &spapr->phbs, list) {
1189 ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1191 error_report("couldn't setup PCI devices in fdt");
1196 spapr_dt_cpus(fdt, spapr);
1198 /* ibm,drc-indexes and friends */
1199 if (smc->dr_lmb_enabled) {
1200 root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1202 if (smc->dr_phb_enabled) {
1203 root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1205 if (mc->nvdimm_supported) {
1206 root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1208 if (root_drc_type_mask) {
1209 _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1212 if (mc->has_hotpluggable_cpus) {
1213 int offset = fdt_path_offset(fdt, "/cpus");
1214 ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1216 error_report("Couldn't set up CPU DR device tree properties");
1221 /* /event-sources */
1222 spapr_dt_events(spapr, fdt);
1225 spapr_dt_rtas(spapr, fdt);
1228 spapr_dt_chosen(spapr, fdt, reset);
1231 if (kvm_enabled()) {
1232 spapr_dt_hypervisor(spapr, fdt);
1235 /* Build memory reserve map */
1237 if (spapr->kernel_size) {
1238 _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1239 spapr->kernel_size)));
1241 if (spapr->initrd_size) {
1242 _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1243 spapr->initrd_size)));
1247 /* NVDIMM devices */
1248 if (mc->nvdimm_supported) {
1249 spapr_dt_persistent_memory(spapr, fdt);
1255 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1257 SpaprMachineState *spapr = opaque;
1259 return (addr & 0x0fffffff) + spapr->kernel_addr;
1262 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1265 CPUPPCState *env = &cpu->env;
1267 /* The TCG path should also be holding the BQL at this point */
1268 g_assert(qemu_mutex_iothread_locked());
1270 g_assert(!vhyp_cpu_in_nested(cpu));
1272 if (FIELD_EX64(env->msr, MSR, PR)) {
1273 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1274 env->gpr[3] = H_PRIVILEGE;
1276 env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1280 struct LPCRSyncState {
1285 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1287 struct LPCRSyncState *s = arg.host_ptr;
1288 PowerPCCPU *cpu = POWERPC_CPU(cs);
1289 CPUPPCState *env = &cpu->env;
1292 cpu_synchronize_state(cs);
1293 lpcr = env->spr[SPR_LPCR];
1296 ppc_store_lpcr(cpu, lpcr);
1299 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1302 struct LPCRSyncState s = {
1307 run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1311 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1312 target_ulong lpid, ppc_v3_pate_t *entry)
1314 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1315 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1317 if (!spapr_cpu->in_nested) {
1320 /* Copy PATE1:GR into PATE0:HR */
1321 entry->dw0 = spapr->patb_entry & PATE0_HR;
1322 entry->dw1 = spapr->patb_entry;
1325 uint64_t patb, pats;
1329 patb = spapr->nested_ptcr & PTCR_PATB;
1330 pats = spapr->nested_ptcr & PTCR_PATS;
1332 /* Calculate number of entries */
1333 pats = 1ull << (pats + 12 - 4);
1340 entry->dw0 = ldq_phys(CPU(cpu)->as, patb);
1341 entry->dw1 = ldq_phys(CPU(cpu)->as, patb + 8);
1347 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1348 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1349 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1350 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1351 #define DIRTY_HPTE(_hpte) ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1354 * Get the fd to access the kernel htab, re-opening it if necessary
1356 static int get_htab_fd(SpaprMachineState *spapr)
1358 Error *local_err = NULL;
1360 if (spapr->htab_fd >= 0) {
1361 return spapr->htab_fd;
1364 spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1365 if (spapr->htab_fd < 0) {
1366 error_report_err(local_err);
1369 return spapr->htab_fd;
1372 void close_htab_fd(SpaprMachineState *spapr)
1374 if (spapr->htab_fd >= 0) {
1375 close(spapr->htab_fd);
1377 spapr->htab_fd = -1;
1380 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1382 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1384 return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1387 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1389 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1391 assert(kvm_enabled());
1397 return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1400 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1403 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1404 hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1408 * HTAB is controlled by KVM. Fetch into temporary buffer
1410 ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1411 kvmppc_read_hptes(hptes, ptex, n);
1416 * HTAB is controlled by QEMU. Just point to the internally
1419 return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1422 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1423 const ppc_hash_pte64_t *hptes,
1426 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1429 g_free((void *)hptes);
1432 /* Nothing to do for qemu managed HPT */
1435 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1436 uint64_t pte0, uint64_t pte1)
1438 SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1439 hwaddr offset = ptex * HASH_PTE_SIZE_64;
1442 kvmppc_write_hpte(ptex, pte0, pte1);
1444 if (pte0 & HPTE64_V_VALID) {
1445 stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1447 * When setting valid, we write PTE1 first. This ensures
1448 * proper synchronization with the reading code in
1449 * ppc_hash64_pteg_search()
1452 stq_p(spapr->htab + offset, pte0);
1454 stq_p(spapr->htab + offset, pte0);
1456 * When clearing it we set PTE0 first. This ensures proper
1457 * synchronization with the reading code in
1458 * ppc_hash64_pteg_search()
1461 stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1466 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1469 hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1470 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1473 /* There should always be a hash table when this is called */
1474 error_report("spapr_hpte_set_c called with no hash table !");
1478 /* The HW performs a non-atomic byte update */
1479 stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1482 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1485 hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1486 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1489 /* There should always be a hash table when this is called */
1490 error_report("spapr_hpte_set_r called with no hash table !");
1494 /* The HW performs a non-atomic byte update */
1495 stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1498 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1502 /* We aim for a hash table of size 1/128 the size of RAM (rounded
1503 * up). The PAPR recommendation is actually 1/64 of RAM size, but
1504 * that's much more than is needed for Linux guests */
1505 shift = ctz64(pow2ceil(ramsize)) - 7;
1506 shift = MAX(shift, 18); /* Minimum architected size */
1507 shift = MIN(shift, 46); /* Maximum architected size */
1511 void spapr_free_hpt(SpaprMachineState *spapr)
1513 g_free(spapr->htab);
1515 spapr->htab_shift = 0;
1516 close_htab_fd(spapr);
1519 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1524 /* Clean up any HPT info from a previous boot */
1525 spapr_free_hpt(spapr);
1527 rc = kvmppc_reset_htab(shift);
1529 if (rc == -EOPNOTSUPP) {
1530 error_setg(errp, "HPT not supported in nested guests");
1535 /* kernel-side HPT needed, but couldn't allocate one */
1536 error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1538 error_append_hint(errp, "Try smaller maxmem?\n");
1540 } else if (rc > 0) {
1541 /* kernel-side HPT allocated */
1544 "Requested order %d HPT, but kernel allocated order %ld",
1546 error_append_hint(errp, "Try smaller maxmem?\n");
1550 spapr->htab_shift = shift;
1553 /* kernel-side HPT not needed, allocate in userspace instead */
1554 size_t size = 1ULL << shift;
1557 spapr->htab = qemu_memalign(size, size);
1558 memset(spapr->htab, 0, size);
1559 spapr->htab_shift = shift;
1561 for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1562 DIRTY_HPTE(HPTE(spapr->htab, i));
1565 /* We're setting up a hash table, so that means we're not radix */
1566 spapr->patb_entry = 0;
1567 spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1571 void spapr_setup_hpt(SpaprMachineState *spapr)
1575 if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1576 hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1578 uint64_t current_ram_size;
1580 current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1581 hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1583 spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1585 if (kvm_enabled()) {
1586 hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1588 /* Check our RMA fits in the possible VRMA */
1589 if (vrma_limit < spapr->rma_size) {
1590 error_report("Unable to create %" HWADDR_PRIu
1591 "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1592 spapr->rma_size / MiB, vrma_limit / MiB);
1598 void spapr_check_mmu_mode(bool guest_radix)
1601 if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1602 error_report("Guest requested unavailable MMU mode (radix).");
1606 if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1607 && !kvmppc_has_cap_mmu_hash_v3()) {
1608 error_report("Guest requested unavailable MMU mode (hash).");
1614 static void spapr_machine_reset(MachineState *machine)
1616 SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1617 PowerPCCPU *first_ppc_cpu;
1622 pef_kvm_reset(machine->cgs, &error_fatal);
1623 spapr_caps_apply(spapr);
1625 first_ppc_cpu = POWERPC_CPU(first_cpu);
1626 if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1627 ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1628 spapr->max_compat_pvr)) {
1630 * If using KVM with radix mode available, VCPUs can be started
1631 * without a HPT because KVM will start them in radix mode.
1632 * Set the GR bit in PATE so that we know there is no HPT.
1634 spapr->patb_entry = PATE1_GR;
1635 spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1637 spapr_setup_hpt(spapr);
1640 qemu_devices_reset();
1642 spapr_ovec_cleanup(spapr->ov5_cas);
1643 spapr->ov5_cas = spapr_ovec_new();
1645 ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1648 * This is fixing some of the default configuration of the XIVE
1649 * devices. To be called after the reset of the machine devices.
1651 spapr_irq_reset(spapr, &error_fatal);
1654 * There is no CAS under qtest. Simulate one to please the code that
1655 * depends on spapr->ov5_cas. This is especially needed to test device
1656 * unplug, so we do that before resetting the DRCs.
1658 if (qtest_enabled()) {
1659 spapr_ovec_cleanup(spapr->ov5_cas);
1660 spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1663 spapr_nvdimm_finish_flushes();
1665 /* DRC reset may cause a device to be unplugged. This will cause troubles
1666 * if this device is used by another device (eg, a running vhost backend
1667 * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1668 * situations, we reset DRCs after all devices have been reset.
1670 spapr_drc_reset_all(spapr);
1672 spapr_clear_pending_events(spapr);
1675 * We place the device tree just below either the top of the RMA,
1676 * or just below 2GB, whichever is lower, so that it can be
1677 * processed with 32-bit real mode code if necessary
1679 fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1681 fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1683 spapr_vof_reset(spapr, fdt, &error_fatal);
1685 * Do not pack the FDT as the client may change properties.
1686 * VOF client does not expect the FDT so we do not load it to the VM.
1690 /* Should only fail if we've built a corrupted tree */
1693 spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1695 cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1697 qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1699 g_free(spapr->fdt_blob);
1700 spapr->fdt_size = fdt_totalsize(fdt);
1701 spapr->fdt_initial_size = spapr->fdt_size;
1702 spapr->fdt_blob = fdt;
1704 /* Set up the entry state */
1705 first_ppc_cpu->env.gpr[5] = 0;
1707 spapr->fwnmi_system_reset_addr = -1;
1708 spapr->fwnmi_machine_check_addr = -1;
1709 spapr->fwnmi_machine_check_interlock = -1;
1711 /* Signal all vCPUs waiting on this condition */
1712 qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1714 migrate_del_blocker(spapr->fwnmi_migration_blocker);
1717 static void spapr_create_nvram(SpaprMachineState *spapr)
1719 DeviceState *dev = qdev_new("spapr-nvram");
1720 DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1723 qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1727 qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1729 spapr->nvram = (struct SpaprNvram *)dev;
1732 static void spapr_rtc_create(SpaprMachineState *spapr)
1734 object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1735 sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1736 &error_fatal, NULL);
1737 qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1738 object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1742 /* Returns whether we want to use VGA or not */
1743 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1745 vga_interface_created = true;
1746 switch (vga_interface_type) {
1754 return pci_vga_init(pci_bus) != NULL;
1757 "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1762 static int spapr_pre_load(void *opaque)
1766 rc = spapr_caps_pre_load(opaque);
1774 static int spapr_post_load(void *opaque, int version_id)
1776 SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1779 err = spapr_caps_post_migration(spapr);
1785 * In earlier versions, there was no separate qdev for the PAPR
1786 * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1787 * So when migrating from those versions, poke the incoming offset
1788 * value into the RTC device
1790 if (version_id < 3) {
1791 err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1797 if (kvm_enabled() && spapr->patb_entry) {
1798 PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1799 bool radix = !!(spapr->patb_entry & PATE1_GR);
1800 bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1803 * Update LPCR:HR and UPRT as they may not be set properly in
1806 spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1807 LPCR_HR | LPCR_UPRT);
1809 err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1811 error_report("Process table config unsupported by the host");
1816 err = spapr_irq_post_load(spapr, version_id);
1824 static int spapr_pre_save(void *opaque)
1828 rc = spapr_caps_pre_save(opaque);
1836 static bool version_before_3(void *opaque, int version_id)
1838 return version_id < 3;
1841 static bool spapr_pending_events_needed(void *opaque)
1843 SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1844 return !QTAILQ_EMPTY(&spapr->pending_events);
1847 static const VMStateDescription vmstate_spapr_event_entry = {
1848 .name = "spapr_event_log_entry",
1850 .minimum_version_id = 1,
1851 .fields = (VMStateField[]) {
1852 VMSTATE_UINT32(summary, SpaprEventLogEntry),
1853 VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1854 VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1855 NULL, extended_length),
1856 VMSTATE_END_OF_LIST()
1860 static const VMStateDescription vmstate_spapr_pending_events = {
1861 .name = "spapr_pending_events",
1863 .minimum_version_id = 1,
1864 .needed = spapr_pending_events_needed,
1865 .fields = (VMStateField[]) {
1866 VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1867 vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1868 VMSTATE_END_OF_LIST()
1872 static bool spapr_ov5_cas_needed(void *opaque)
1874 SpaprMachineState *spapr = opaque;
1875 SpaprOptionVector *ov5_mask = spapr_ovec_new();
1878 /* Prior to the introduction of SpaprOptionVector, we had two option
1879 * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1880 * Both of these options encode machine topology into the device-tree
1881 * in such a way that the now-booted OS should still be able to interact
1882 * appropriately with QEMU regardless of what options were actually
1883 * negotiatied on the source side.
1885 * As such, we can avoid migrating the CAS-negotiated options if these
1886 * are the only options available on the current machine/platform.
1887 * Since these are the only options available for pseries-2.7 and
1888 * earlier, this allows us to maintain old->new/new->old migration
1891 * For QEMU 2.8+, there are additional CAS-negotiatable options available
1892 * via default pseries-2.8 machines and explicit command-line parameters.
1893 * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1894 * of the actual CAS-negotiated values to continue working properly. For
1895 * example, availability of memory unplug depends on knowing whether
1896 * OV5_HP_EVT was negotiated via CAS.
1898 * Thus, for any cases where the set of available CAS-negotiatable
1899 * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1900 * include the CAS-negotiated options in the migration stream, unless
1901 * if they affect boot time behaviour only.
1903 spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1904 spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1905 spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1907 /* We need extra information if we have any bits outside the mask
1909 cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1911 spapr_ovec_cleanup(ov5_mask);
1916 static const VMStateDescription vmstate_spapr_ov5_cas = {
1917 .name = "spapr_option_vector_ov5_cas",
1919 .minimum_version_id = 1,
1920 .needed = spapr_ov5_cas_needed,
1921 .fields = (VMStateField[]) {
1922 VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1923 vmstate_spapr_ovec, SpaprOptionVector),
1924 VMSTATE_END_OF_LIST()
1928 static bool spapr_patb_entry_needed(void *opaque)
1930 SpaprMachineState *spapr = opaque;
1932 return !!spapr->patb_entry;
1935 static const VMStateDescription vmstate_spapr_patb_entry = {
1936 .name = "spapr_patb_entry",
1938 .minimum_version_id = 1,
1939 .needed = spapr_patb_entry_needed,
1940 .fields = (VMStateField[]) {
1941 VMSTATE_UINT64(patb_entry, SpaprMachineState),
1942 VMSTATE_END_OF_LIST()
1946 static bool spapr_irq_map_needed(void *opaque)
1948 SpaprMachineState *spapr = opaque;
1950 return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1953 static const VMStateDescription vmstate_spapr_irq_map = {
1954 .name = "spapr_irq_map",
1956 .minimum_version_id = 1,
1957 .needed = spapr_irq_map_needed,
1958 .fields = (VMStateField[]) {
1959 VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1960 VMSTATE_END_OF_LIST()
1964 static bool spapr_dtb_needed(void *opaque)
1966 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1968 return smc->update_dt_enabled;
1971 static int spapr_dtb_pre_load(void *opaque)
1973 SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1975 g_free(spapr->fdt_blob);
1976 spapr->fdt_blob = NULL;
1977 spapr->fdt_size = 0;
1982 static const VMStateDescription vmstate_spapr_dtb = {
1983 .name = "spapr_dtb",
1985 .minimum_version_id = 1,
1986 .needed = spapr_dtb_needed,
1987 .pre_load = spapr_dtb_pre_load,
1988 .fields = (VMStateField[]) {
1989 VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1990 VMSTATE_UINT32(fdt_size, SpaprMachineState),
1991 VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1993 VMSTATE_END_OF_LIST()
1997 static bool spapr_fwnmi_needed(void *opaque)
1999 SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2001 return spapr->fwnmi_machine_check_addr != -1;
2004 static int spapr_fwnmi_pre_save(void *opaque)
2006 SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2009 * Check if machine check handling is in progress and print a
2012 if (spapr->fwnmi_machine_check_interlock != -1) {
2013 warn_report("A machine check is being handled during migration. The"
2014 "handler may run and log hardware error on the destination");
2020 static const VMStateDescription vmstate_spapr_fwnmi = {
2021 .name = "spapr_fwnmi",
2023 .minimum_version_id = 1,
2024 .needed = spapr_fwnmi_needed,
2025 .pre_save = spapr_fwnmi_pre_save,
2026 .fields = (VMStateField[]) {
2027 VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2028 VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2029 VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2030 VMSTATE_END_OF_LIST()
2034 static const VMStateDescription vmstate_spapr = {
2037 .minimum_version_id = 1,
2038 .pre_load = spapr_pre_load,
2039 .post_load = spapr_post_load,
2040 .pre_save = spapr_pre_save,
2041 .fields = (VMStateField[]) {
2042 /* used to be @next_irq */
2043 VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2046 VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2048 VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2049 VMSTATE_END_OF_LIST()
2051 .subsections = (const VMStateDescription*[]) {
2052 &vmstate_spapr_ov5_cas,
2053 &vmstate_spapr_patb_entry,
2054 &vmstate_spapr_pending_events,
2055 &vmstate_spapr_cap_htm,
2056 &vmstate_spapr_cap_vsx,
2057 &vmstate_spapr_cap_dfp,
2058 &vmstate_spapr_cap_cfpc,
2059 &vmstate_spapr_cap_sbbc,
2060 &vmstate_spapr_cap_ibs,
2061 &vmstate_spapr_cap_hpt_maxpagesize,
2062 &vmstate_spapr_irq_map,
2063 &vmstate_spapr_cap_nested_kvm_hv,
2065 &vmstate_spapr_cap_large_decr,
2066 &vmstate_spapr_cap_ccf_assist,
2067 &vmstate_spapr_cap_fwnmi,
2068 &vmstate_spapr_fwnmi,
2069 &vmstate_spapr_cap_rpt_invalidate,
2074 static int htab_save_setup(QEMUFile *f, void *opaque)
2076 SpaprMachineState *spapr = opaque;
2078 /* "Iteration" header */
2079 if (!spapr->htab_shift) {
2080 qemu_put_be32(f, -1);
2082 qemu_put_be32(f, spapr->htab_shift);
2086 spapr->htab_save_index = 0;
2087 spapr->htab_first_pass = true;
2089 if (spapr->htab_shift) {
2090 assert(kvm_enabled());
2098 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2099 int chunkstart, int n_valid, int n_invalid)
2101 qemu_put_be32(f, chunkstart);
2102 qemu_put_be16(f, n_valid);
2103 qemu_put_be16(f, n_invalid);
2104 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2105 HASH_PTE_SIZE_64 * n_valid);
2108 static void htab_save_end_marker(QEMUFile *f)
2110 qemu_put_be32(f, 0);
2111 qemu_put_be16(f, 0);
2112 qemu_put_be16(f, 0);
2115 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2118 bool has_timeout = max_ns != -1;
2119 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2120 int index = spapr->htab_save_index;
2121 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2123 assert(spapr->htab_first_pass);
2128 /* Consume invalid HPTEs */
2129 while ((index < htabslots)
2130 && !HPTE_VALID(HPTE(spapr->htab, index))) {
2131 CLEAN_HPTE(HPTE(spapr->htab, index));
2135 /* Consume valid HPTEs */
2137 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2138 && HPTE_VALID(HPTE(spapr->htab, index))) {
2139 CLEAN_HPTE(HPTE(spapr->htab, index));
2143 if (index > chunkstart) {
2144 int n_valid = index - chunkstart;
2146 htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2149 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2153 } while ((index < htabslots) && !qemu_file_rate_limit(f));
2155 if (index >= htabslots) {
2156 assert(index == htabslots);
2158 spapr->htab_first_pass = false;
2160 spapr->htab_save_index = index;
2163 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2166 bool final = max_ns < 0;
2167 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2168 int examined = 0, sent = 0;
2169 int index = spapr->htab_save_index;
2170 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2172 assert(!spapr->htab_first_pass);
2175 int chunkstart, invalidstart;
2177 /* Consume non-dirty HPTEs */
2178 while ((index < htabslots)
2179 && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2185 /* Consume valid dirty HPTEs */
2186 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2187 && HPTE_DIRTY(HPTE(spapr->htab, index))
2188 && HPTE_VALID(HPTE(spapr->htab, index))) {
2189 CLEAN_HPTE(HPTE(spapr->htab, index));
2194 invalidstart = index;
2195 /* Consume invalid dirty HPTEs */
2196 while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2197 && HPTE_DIRTY(HPTE(spapr->htab, index))
2198 && !HPTE_VALID(HPTE(spapr->htab, index))) {
2199 CLEAN_HPTE(HPTE(spapr->htab, index));
2204 if (index > chunkstart) {
2205 int n_valid = invalidstart - chunkstart;
2206 int n_invalid = index - invalidstart;
2208 htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2209 sent += index - chunkstart;
2211 if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2216 if (examined >= htabslots) {
2220 if (index >= htabslots) {
2221 assert(index == htabslots);
2224 } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2226 if (index >= htabslots) {
2227 assert(index == htabslots);
2231 spapr->htab_save_index = index;
2233 return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2236 #define MAX_ITERATION_NS 5000000 /* 5 ms */
2237 #define MAX_KVM_BUF_SIZE 2048
2239 static int htab_save_iterate(QEMUFile *f, void *opaque)
2241 SpaprMachineState *spapr = opaque;
2245 /* Iteration header */
2246 if (!spapr->htab_shift) {
2247 qemu_put_be32(f, -1);
2250 qemu_put_be32(f, 0);
2254 assert(kvm_enabled());
2256 fd = get_htab_fd(spapr);
2261 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2265 } else if (spapr->htab_first_pass) {
2266 htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2268 rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2271 htab_save_end_marker(f);
2276 static int htab_save_complete(QEMUFile *f, void *opaque)
2278 SpaprMachineState *spapr = opaque;
2281 /* Iteration header */
2282 if (!spapr->htab_shift) {
2283 qemu_put_be32(f, -1);
2286 qemu_put_be32(f, 0);
2292 assert(kvm_enabled());
2294 fd = get_htab_fd(spapr);
2299 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2304 if (spapr->htab_first_pass) {
2305 htab_save_first_pass(f, spapr, -1);
2307 htab_save_later_pass(f, spapr, -1);
2311 htab_save_end_marker(f);
2316 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2318 SpaprMachineState *spapr = opaque;
2319 uint32_t section_hdr;
2321 Error *local_err = NULL;
2323 if (version_id < 1 || version_id > 1) {
2324 error_report("htab_load() bad version");
2328 section_hdr = qemu_get_be32(f);
2330 if (section_hdr == -1) {
2331 spapr_free_hpt(spapr);
2338 /* First section gives the htab size */
2339 ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2341 error_report_err(local_err);
2348 assert(kvm_enabled());
2350 fd = kvmppc_get_htab_fd(true, 0, &local_err);
2352 error_report_err(local_err);
2359 uint16_t n_valid, n_invalid;
2361 index = qemu_get_be32(f);
2362 n_valid = qemu_get_be16(f);
2363 n_invalid = qemu_get_be16(f);
2365 if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2370 if ((index + n_valid + n_invalid) >
2371 (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2372 /* Bad index in stream */
2374 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2375 index, n_valid, n_invalid, spapr->htab_shift);
2381 qemu_get_buffer(f, HPTE(spapr->htab, index),
2382 HASH_PTE_SIZE_64 * n_valid);
2385 memset(HPTE(spapr->htab, index + n_valid), 0,
2386 HASH_PTE_SIZE_64 * n_invalid);
2393 rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2396 error_report_err(local_err);
2410 static void htab_save_cleanup(void *opaque)
2412 SpaprMachineState *spapr = opaque;
2414 close_htab_fd(spapr);
2417 static SaveVMHandlers savevm_htab_handlers = {
2418 .save_setup = htab_save_setup,
2419 .save_live_iterate = htab_save_iterate,
2420 .save_live_complete_precopy = htab_save_complete,
2421 .save_cleanup = htab_save_cleanup,
2422 .load_state = htab_load,
2425 static void spapr_boot_set(void *opaque, const char *boot_device,
2428 SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2430 g_free(spapr->boot_device);
2431 spapr->boot_device = g_strdup(boot_device);
2434 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2436 MachineState *machine = MACHINE(spapr);
2437 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2438 uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2441 for (i = 0; i < nr_lmbs; i++) {
2444 addr = i * lmb_size + machine->device_memory->base;
2445 spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2451 * If RAM size, maxmem size and individual node mem sizes aren't aligned
2452 * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2453 * since we can't support such unaligned sizes with DRCONF_MEMORY.
2455 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2459 if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2460 error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2461 " is not aligned to %" PRIu64 " MiB",
2463 SPAPR_MEMORY_BLOCK_SIZE / MiB);
2467 if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2468 error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2469 " is not aligned to %" PRIu64 " MiB",
2471 SPAPR_MEMORY_BLOCK_SIZE / MiB);
2475 for (i = 0; i < machine->numa_state->num_nodes; i++) {
2476 if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2478 "Node %d memory size 0x%" PRIx64
2479 " is not aligned to %" PRIu64 " MiB",
2480 i, machine->numa_state->nodes[i].node_mem,
2481 SPAPR_MEMORY_BLOCK_SIZE / MiB);
2487 /* find cpu slot in machine->possible_cpus by core_id */
2488 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2490 int index = id / ms->smp.threads;
2492 if (index >= ms->possible_cpus->len) {
2498 return &ms->possible_cpus->cpus[index];
2501 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2503 MachineState *ms = MACHINE(spapr);
2504 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2505 Error *local_err = NULL;
2506 bool vsmt_user = !!spapr->vsmt;
2507 int kvm_smt = kvmppc_smt_threads();
2509 unsigned int smp_threads = ms->smp.threads;
2511 if (!kvm_enabled() && (smp_threads > 1)) {
2512 error_setg(errp, "TCG cannot support more than 1 thread/core "
2513 "on a pseries machine");
2516 if (!is_power_of_2(smp_threads)) {
2517 error_setg(errp, "Cannot support %d threads/core on a pseries "
2518 "machine because it must be a power of 2", smp_threads);
2522 /* Detemine the VSMT mode to use: */
2524 if (spapr->vsmt < smp_threads) {
2525 error_setg(errp, "Cannot support VSMT mode %d"
2526 " because it must be >= threads/core (%d)",
2527 spapr->vsmt, smp_threads);
2530 /* In this case, spapr->vsmt has been set by the command line */
2531 } else if (!smc->smp_threads_vsmt) {
2533 * Default VSMT value is tricky, because we need it to be as
2534 * consistent as possible (for migration), but this requires
2535 * changing it for at least some existing cases. We pick 8 as
2536 * the value that we'd get with KVM on POWER8, the
2537 * overwhelmingly common case in production systems.
2539 spapr->vsmt = MAX(8, smp_threads);
2541 spapr->vsmt = smp_threads;
2544 /* KVM: If necessary, set the SMT mode: */
2545 if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2546 ret = kvmppc_set_smt_threads(spapr->vsmt);
2548 /* Looks like KVM isn't able to change VSMT mode */
2549 error_setg(&local_err,
2550 "Failed to set KVM's VSMT mode to %d (errno %d)",
2552 /* We can live with that if the default one is big enough
2553 * for the number of threads, and a submultiple of the one
2554 * we want. In this case we'll waste some vcpu ids, but
2555 * behaviour will be correct */
2556 if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2557 warn_report_err(local_err);
2560 error_append_hint(&local_err,
2561 "On PPC, a VM with %d threads/core"
2562 " on a host with %d threads/core"
2563 " requires the use of VSMT mode %d.\n",
2564 smp_threads, kvm_smt, spapr->vsmt);
2566 kvmppc_error_append_smt_possible_hint(&local_err);
2567 error_propagate(errp, local_err);
2571 /* else TCG: nothing to do currently */
2574 static void spapr_init_cpus(SpaprMachineState *spapr)
2576 MachineState *machine = MACHINE(spapr);
2577 MachineClass *mc = MACHINE_GET_CLASS(machine);
2578 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2579 const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2580 const CPUArchIdList *possible_cpus;
2581 unsigned int smp_cpus = machine->smp.cpus;
2582 unsigned int smp_threads = machine->smp.threads;
2583 unsigned int max_cpus = machine->smp.max_cpus;
2584 int boot_cores_nr = smp_cpus / smp_threads;
2587 possible_cpus = mc->possible_cpu_arch_ids(machine);
2588 if (mc->has_hotpluggable_cpus) {
2589 if (smp_cpus % smp_threads) {
2590 error_report("smp_cpus (%u) must be multiple of threads (%u)",
2591 smp_cpus, smp_threads);
2594 if (max_cpus % smp_threads) {
2595 error_report("max_cpus (%u) must be multiple of threads (%u)",
2596 max_cpus, smp_threads);
2600 if (max_cpus != smp_cpus) {
2601 error_report("This machine version does not support CPU hotplug");
2604 boot_cores_nr = possible_cpus->len;
2607 if (smc->pre_2_10_has_unused_icps) {
2610 for (i = 0; i < spapr_max_server_number(spapr); i++) {
2611 /* Dummy entries get deregistered when real ICPState objects
2612 * are registered during CPU core hotplug.
2614 pre_2_10_vmstate_register_dummy_icp(i);
2618 for (i = 0; i < possible_cpus->len; i++) {
2619 int core_id = i * smp_threads;
2621 if (mc->has_hotpluggable_cpus) {
2622 spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2623 spapr_vcpu_id(spapr, core_id));
2626 if (i < boot_cores_nr) {
2627 Object *core = object_new(type);
2628 int nr_threads = smp_threads;
2630 /* Handle the partially filled core for older machine types */
2631 if ((i + 1) * smp_threads >= smp_cpus) {
2632 nr_threads = smp_cpus - i * smp_threads;
2635 object_property_set_int(core, "nr-threads", nr_threads,
2637 object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2639 qdev_realize(DEVICE(core), NULL, &error_fatal);
2646 static PCIHostState *spapr_create_default_phb(void)
2650 dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2651 qdev_prop_set_uint32(dev, "index", 0);
2652 sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2654 return PCI_HOST_BRIDGE(dev);
2657 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2659 MachineState *machine = MACHINE(spapr);
2660 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2661 hwaddr rma_size = machine->ram_size;
2662 hwaddr node0_size = spapr_node0_size(machine);
2664 /* RMA has to fit in the first NUMA node */
2665 rma_size = MIN(rma_size, node0_size);
2668 * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2671 rma_size = MIN(rma_size, 1 * TiB);
2674 * Clamp the RMA size based on machine type. This is for
2675 * migration compatibility with older qemu versions, which limited
2676 * the RMA size for complicated and mostly bad reasons.
2678 if (smc->rma_limit) {
2679 rma_size = MIN(rma_size, smc->rma_limit);
2682 if (rma_size < MIN_RMA_SLOF) {
2684 "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2685 "ldMiB guest RMA (Real Mode Area memory)",
2686 MIN_RMA_SLOF / MiB);
2693 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2695 MachineState *machine = MACHINE(spapr);
2698 for (i = 0; i < machine->ram_slots; i++) {
2699 spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2703 /* pSeries LPAR / sPAPR hardware init */
2704 static void spapr_machine_init(MachineState *machine)
2706 SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2707 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2708 MachineClass *mc = MACHINE_GET_CLASS(machine);
2709 const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2710 const char *bios_name = machine->firmware ?: bios_default;
2711 g_autofree char *filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2712 const char *kernel_filename = machine->kernel_filename;
2713 const char *initrd_filename = machine->initrd_filename;
2717 MemoryRegion *sysmem = get_system_memory();
2718 long load_limit, fw_size;
2719 Error *resize_hpt_err = NULL;
2722 error_report("Could not find LPAR firmware '%s'", bios_name);
2725 fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2727 error_report("Could not load LPAR firmware '%s'", filename);
2732 * if Secure VM (PEF) support is configured, then initialize it
2734 pef_kvm_init(machine->cgs, &error_fatal);
2736 msi_nonbroken = true;
2738 QLIST_INIT(&spapr->phbs);
2739 QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2741 /* Determine capabilities to run with */
2742 spapr_caps_init(spapr);
2744 kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2745 if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2747 * If the user explicitly requested a mode we should either
2748 * supply it, or fail completely (which we do below). But if
2749 * it's not set explicitly, we reset our mode to something
2752 if (resize_hpt_err) {
2753 spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2754 error_free(resize_hpt_err);
2755 resize_hpt_err = NULL;
2757 spapr->resize_hpt = smc->resize_hpt_default;
2761 assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2763 if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2765 * User requested HPT resize, but this host can't supply it. Bail out
2767 error_report_err(resize_hpt_err);
2770 error_free(resize_hpt_err);
2772 spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2774 /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2775 load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2778 * VSMT must be set in order to be able to compute VCPU ids, ie to
2779 * call spapr_max_server_number() or spapr_vcpu_id().
2781 spapr_set_vsmt_mode(spapr, &error_fatal);
2783 /* Set up Interrupt Controller before we create the VCPUs */
2784 spapr_irq_init(spapr, &error_fatal);
2786 /* Set up containers for ibm,client-architecture-support negotiated options
2788 spapr->ov5 = spapr_ovec_new();
2789 spapr->ov5_cas = spapr_ovec_new();
2791 if (smc->dr_lmb_enabled) {
2792 spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2793 spapr_validate_node_memory(machine, &error_fatal);
2796 spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2798 /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2799 if (!smc->pre_6_2_numa_affinity) {
2800 spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2803 /* advertise support for dedicated HP event source to guests */
2804 if (spapr->use_hotplug_event_source) {
2805 spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2808 /* advertise support for HPT resizing */
2809 if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2810 spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2813 /* advertise support for ibm,dyamic-memory-v2 */
2814 spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2816 /* advertise XIVE on POWER9 machines */
2817 if (spapr->irq->xive) {
2818 spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2822 spapr_init_cpus(spapr);
2824 spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine);
2826 /* Init numa_assoc_array */
2827 spapr_numa_associativity_init(spapr, machine);
2829 if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2830 ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2831 spapr->max_compat_pvr)) {
2832 spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2833 /* KVM and TCG always allow GTSE with radix... */
2834 spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2836 /* ... but not with hash (currently). */
2838 if (kvm_enabled()) {
2839 /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2840 kvmppc_enable_logical_ci_hcalls();
2841 kvmppc_enable_set_mode_hcall();
2843 /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2844 kvmppc_enable_clear_ref_mod_hcalls();
2846 /* Enable H_PAGE_INIT */
2847 kvmppc_enable_h_page_init();
2851 memory_region_add_subregion(sysmem, 0, machine->ram);
2853 /* always allocate the device memory information */
2854 machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2856 /* initialize hotplug memory address space */
2857 if (machine->ram_size < machine->maxram_size) {
2858 ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2860 * Limit the number of hotpluggable memory slots to half the number
2861 * slots that KVM supports, leaving the other half for PCI and other
2862 * devices. However ensure that number of slots doesn't drop below 32.
2864 int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2865 SPAPR_MAX_RAM_SLOTS;
2867 if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2868 max_memslots = SPAPR_MAX_RAM_SLOTS;
2870 if (machine->ram_slots > max_memslots) {
2871 error_report("Specified number of memory slots %"
2872 PRIu64" exceeds max supported %d",
2873 machine->ram_slots, max_memslots);
2877 machine->device_memory->base = ROUND_UP(machine->ram_size,
2878 SPAPR_DEVICE_MEM_ALIGN);
2879 memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2880 "device-memory", device_mem_size);
2881 memory_region_add_subregion(sysmem, machine->device_memory->base,
2882 &machine->device_memory->mr);
2885 if (smc->dr_lmb_enabled) {
2886 spapr_create_lmb_dr_connectors(spapr);
2889 if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2890 /* Create the error string for live migration blocker */
2891 error_setg(&spapr->fwnmi_migration_blocker,
2892 "A machine check is being handled during migration. The handler"
2893 "may run and log hardware error on the destination");
2896 if (mc->nvdimm_supported) {
2897 spapr_create_nvdimm_dr_connectors(spapr);
2900 /* Set up RTAS event infrastructure */
2901 spapr_events_init(spapr);
2903 /* Set up the RTC RTAS interfaces */
2904 spapr_rtc_create(spapr);
2906 /* Set up VIO bus */
2907 spapr->vio_bus = spapr_vio_bus_init();
2909 for (i = 0; serial_hd(i); i++) {
2910 spapr_vty_create(spapr->vio_bus, serial_hd(i));
2913 /* We always have at least the nvram device on VIO */
2914 spapr_create_nvram(spapr);
2917 * Setup hotplug / dynamic-reconfiguration connectors. top-level
2918 * connectors (described in root DT node's "ibm,drc-types" property)
2919 * are pre-initialized here. additional child connectors (such as
2920 * connectors for a PHBs PCI slots) are added as needed during their
2921 * parent's realization.
2923 if (smc->dr_phb_enabled) {
2924 for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2925 spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2930 spapr_pci_rtas_init();
2932 phb = spapr_create_default_phb();
2934 for (i = 0; i < nb_nics; i++) {
2935 NICInfo *nd = &nd_table[i];
2938 nd->model = g_strdup("spapr-vlan");
2941 if (g_str_equal(nd->model, "spapr-vlan") ||
2942 g_str_equal(nd->model, "ibmveth")) {
2943 spapr_vlan_create(spapr->vio_bus, nd);
2945 pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2949 for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2950 spapr_vscsi_create(spapr->vio_bus);
2954 has_vga = spapr_vga_init(phb->bus, &error_fatal);
2956 spapr->want_stdout_path = !machine->enable_graphics;
2957 machine->usb |= defaults_enabled() && !machine->usb_disabled;
2959 spapr->want_stdout_path = true;
2963 if (smc->use_ohci_by_default) {
2964 pci_create_simple(phb->bus, -1, "pci-ohci");
2966 pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2970 USBBus *usb_bus = usb_bus_find(-1);
2972 usb_create_simple(usb_bus, "usb-kbd");
2973 usb_create_simple(usb_bus, "usb-mouse");
2977 if (kernel_filename) {
2978 spapr->kernel_size = load_elf(kernel_filename, NULL,
2979 translate_kernel_address, spapr,
2980 NULL, NULL, NULL, NULL, 1,
2981 PPC_ELF_MACHINE, 0, 0);
2982 if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2983 spapr->kernel_size = load_elf(kernel_filename, NULL,
2984 translate_kernel_address, spapr,
2985 NULL, NULL, NULL, NULL, 0,
2986 PPC_ELF_MACHINE, 0, 0);
2987 spapr->kernel_le = spapr->kernel_size > 0;
2989 if (spapr->kernel_size < 0) {
2990 error_report("error loading %s: %s", kernel_filename,
2991 load_elf_strerror(spapr->kernel_size));
2996 if (initrd_filename) {
2997 /* Try to locate the initrd in the gap between the kernel
2998 * and the firmware. Add a bit of space just in case
3000 spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3001 + 0x1ffff) & ~0xffff;
3002 spapr->initrd_size = load_image_targphys(initrd_filename,
3005 - spapr->initrd_base);
3006 if (spapr->initrd_size < 0) {
3007 error_report("could not load initial ram disk '%s'",
3014 /* FIXME: Should register things through the MachineState's qdev
3015 * interface, this is a legacy from the sPAPREnvironment structure
3016 * which predated MachineState but had a similar function */
3017 vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3018 register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3019 &savevm_htab_handlers, spapr);
3021 qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3023 qemu_register_boot_set(spapr_boot_set, spapr);
3026 * Nothing needs to be done to resume a suspended guest because
3027 * suspending does not change the machine state, so no need for
3028 * a ->wakeup method.
3030 qemu_register_wakeup_support();
3032 if (kvm_enabled()) {
3033 /* to stop and start vmclock */
3034 qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3037 kvmppc_spapr_enable_inkernel_multitce();
3040 qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3042 spapr->vof->fw_size = fw_size; /* for claim() on itself */
3043 spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3047 #define DEFAULT_KVM_TYPE "auto"
3048 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3051 * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3052 * accomodate the 'HV' and 'PV' formats that exists in the
3053 * wild. The 'auto' mode is being introduced already as
3054 * lower-case, thus we don't need to bother checking for
3057 if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3061 if (!g_ascii_strcasecmp(vm_type, "hv")) {
3065 if (!g_ascii_strcasecmp(vm_type, "pr")) {
3069 error_report("Unknown kvm-type specified '%s'", vm_type);
3074 * Implementation of an interface to adjust firmware path
3075 * for the bootindex property handling.
3077 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3080 #define CAST(type, obj, name) \
3081 ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3082 SCSIDevice *d = CAST(SCSIDevice, dev, TYPE_SCSI_DEVICE);
3083 SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3084 VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3085 PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3088 void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3089 VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3090 USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3094 * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3095 * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3096 * 0x8000 | (target << 8) | (bus << 5) | lun
3097 * (see the "Logical unit addressing format" table in SAM5)
3099 unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3100 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3101 (uint64_t)id << 48);
3102 } else if (virtio) {
3104 * We use SRP luns of the form 01000000 | (target << 8) | lun
3105 * in the top 32 bits of the 64-bit LUN
3106 * Note: the quote above is from SLOF and it is wrong,
3107 * the actual binding is:
3108 * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3110 unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3111 if (d->lun >= 256) {
3112 /* Use the LUN "flat space addressing method" */
3115 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3116 (uint64_t)id << 32);
3119 * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3120 * in the top 32 bits of the 64-bit LUN
3122 unsigned usb_port = atoi(usb->port->path);
3123 unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3124 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3125 (uint64_t)id << 32);
3130 * SLOF probes the USB devices, and if it recognizes that the device is a
3131 * storage device, it changes its name to "storage" instead of "usb-host",
3132 * and additionally adds a child node for the SCSI LUN, so the correct
3133 * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3135 if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3136 USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3137 if (usb_device_is_scsi_storage(usbdev)) {
3138 return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3143 /* Replace "pci" with "pci@800000020000000" */
3144 return g_strdup_printf("pci@%"PRIX64, phb->buid);
3148 /* Same logic as virtio above */
3149 unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3150 return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3153 if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3154 /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3155 PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3156 return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3160 return spapr_pci_fw_dev_name(pcidev);
3166 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3168 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3170 return g_strdup(spapr->kvm_type);
3173 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3175 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3177 g_free(spapr->kvm_type);
3178 spapr->kvm_type = g_strdup(value);
3181 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3183 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3185 return spapr->use_hotplug_event_source;
3188 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3191 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3193 spapr->use_hotplug_event_source = value;
3196 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3201 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3203 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3205 switch (spapr->resize_hpt) {
3206 case SPAPR_RESIZE_HPT_DEFAULT:
3207 return g_strdup("default");
3208 case SPAPR_RESIZE_HPT_DISABLED:
3209 return g_strdup("disabled");
3210 case SPAPR_RESIZE_HPT_ENABLED:
3211 return g_strdup("enabled");
3212 case SPAPR_RESIZE_HPT_REQUIRED:
3213 return g_strdup("required");
3215 g_assert_not_reached();
3218 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3220 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3222 if (strcmp(value, "default") == 0) {
3223 spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3224 } else if (strcmp(value, "disabled") == 0) {
3225 spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3226 } else if (strcmp(value, "enabled") == 0) {
3227 spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3228 } else if (strcmp(value, "required") == 0) {
3229 spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3231 error_setg(errp, "Bad value for \"resize-hpt\" property");
3235 static bool spapr_get_vof(Object *obj, Error **errp)
3237 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3239 return spapr->vof != NULL;
3242 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3244 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3247 vof_cleanup(spapr->vof);
3254 spapr->vof = g_malloc0(sizeof(*spapr->vof));
3257 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3259 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3261 if (spapr->irq == &spapr_irq_xics_legacy) {
3262 return g_strdup("legacy");
3263 } else if (spapr->irq == &spapr_irq_xics) {
3264 return g_strdup("xics");
3265 } else if (spapr->irq == &spapr_irq_xive) {
3266 return g_strdup("xive");
3267 } else if (spapr->irq == &spapr_irq_dual) {
3268 return g_strdup("dual");
3270 g_assert_not_reached();
3273 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3275 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3277 if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3278 error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3282 /* The legacy IRQ backend can not be set */
3283 if (strcmp(value, "xics") == 0) {
3284 spapr->irq = &spapr_irq_xics;
3285 } else if (strcmp(value, "xive") == 0) {
3286 spapr->irq = &spapr_irq_xive;
3287 } else if (strcmp(value, "dual") == 0) {
3288 spapr->irq = &spapr_irq_dual;
3290 error_setg(errp, "Bad value for \"ic-mode\" property");
3294 static char *spapr_get_host_model(Object *obj, Error **errp)
3296 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3298 return g_strdup(spapr->host_model);
3301 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3303 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3305 g_free(spapr->host_model);
3306 spapr->host_model = g_strdup(value);
3309 static char *spapr_get_host_serial(Object *obj, Error **errp)
3311 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3313 return g_strdup(spapr->host_serial);
3316 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3318 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3320 g_free(spapr->host_serial);
3321 spapr->host_serial = g_strdup(value);
3324 static void spapr_instance_init(Object *obj)
3326 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3327 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3328 MachineState *ms = MACHINE(spapr);
3329 MachineClass *mc = MACHINE_GET_CLASS(ms);
3332 * NVDIMM support went live in 5.1 without considering that, in
3333 * other archs, the user needs to enable NVDIMM support with the
3334 * 'nvdimm' machine option and the default behavior is NVDIMM
3335 * support disabled. It is too late to roll back to the standard
3336 * behavior without breaking 5.1 guests.
3338 if (mc->nvdimm_supported) {
3339 ms->nvdimms_state->is_enabled = true;
3342 spapr->htab_fd = -1;
3343 spapr->use_hotplug_event_source = true;
3344 spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3345 object_property_add_str(obj, "kvm-type",
3346 spapr_get_kvm_type, spapr_set_kvm_type);
3347 object_property_set_description(obj, "kvm-type",
3348 "Specifies the KVM virtualization mode (auto,"
3349 " hv, pr). Defaults to 'auto'. This mode will use"
3350 " any available KVM module loaded in the host,"
3351 " where kvm_hv takes precedence if both kvm_hv and"
3352 " kvm_pr are loaded.");
3353 object_property_add_bool(obj, "modern-hotplug-events",
3354 spapr_get_modern_hotplug_events,
3355 spapr_set_modern_hotplug_events);
3356 object_property_set_description(obj, "modern-hotplug-events",
3357 "Use dedicated hotplug event mechanism in"
3358 " place of standard EPOW events when possible"
3359 " (required for memory hot-unplug support)");
3360 ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3361 "Maximum permitted CPU compatibility mode");
3363 object_property_add_str(obj, "resize-hpt",
3364 spapr_get_resize_hpt, spapr_set_resize_hpt);
3365 object_property_set_description(obj, "resize-hpt",
3366 "Resizing of the Hash Page Table (enabled, disabled, required)");
3367 object_property_add_uint32_ptr(obj, "vsmt",
3368 &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3369 object_property_set_description(obj, "vsmt",
3370 "Virtual SMT: KVM behaves as if this were"
3371 " the host's SMT mode");
3373 object_property_add_bool(obj, "vfio-no-msix-emulation",
3374 spapr_get_msix_emulation, NULL);
3376 object_property_add_uint64_ptr(obj, "kernel-addr",
3377 &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3378 object_property_set_description(obj, "kernel-addr",
3379 stringify(KERNEL_LOAD_ADDR)
3380 " for -kernel is the default");
3381 spapr->kernel_addr = KERNEL_LOAD_ADDR;
3383 object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3384 object_property_set_description(obj, "x-vof",
3385 "Enable Virtual Open Firmware (experimental)");
3387 /* The machine class defines the default interrupt controller mode */
3388 spapr->irq = smc->irq;
3389 object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3391 object_property_set_description(obj, "ic-mode",
3392 "Specifies the interrupt controller mode (xics, xive, dual)");
3394 object_property_add_str(obj, "host-model",
3395 spapr_get_host_model, spapr_set_host_model);
3396 object_property_set_description(obj, "host-model",
3397 "Host model to advertise in guest device tree");
3398 object_property_add_str(obj, "host-serial",
3399 spapr_get_host_serial, spapr_set_host_serial);
3400 object_property_set_description(obj, "host-serial",
3401 "Host serial number to advertise in guest device tree");
3404 static void spapr_machine_finalizefn(Object *obj)
3406 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3408 g_free(spapr->kvm_type);
3411 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3413 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3414 PowerPCCPU *cpu = POWERPC_CPU(cs);
3415 CPUPPCState *env = &cpu->env;
3417 cpu_synchronize_state(cs);
3418 /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3419 if (spapr->fwnmi_system_reset_addr != -1) {
3420 uint64_t rtas_addr, addr;
3422 /* get rtas addr from fdt */
3423 rtas_addr = spapr_get_rtas_addr();
3425 qemu_system_guest_panicked(NULL);
3429 addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3430 stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3431 stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3434 ppc_cpu_do_system_reset(cs);
3435 if (spapr->fwnmi_system_reset_addr != -1) {
3436 env->nip = spapr->fwnmi_system_reset_addr;
3440 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3445 async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3449 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3450 void *fdt, int *fdt_start_offset, Error **errp)
3455 addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3456 node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3458 *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3459 SPAPR_MEMORY_BLOCK_SIZE);
3463 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3464 bool dedicated_hp_event_source)
3467 uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3469 uint64_t addr = addr_start;
3470 bool hotplugged = spapr_drc_hotplugged(dev);
3472 for (i = 0; i < nr_lmbs; i++) {
3473 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3474 addr / SPAPR_MEMORY_BLOCK_SIZE);
3478 * memory_device_get_free_addr() provided a range of free addresses
3479 * that doesn't overlap with any existing mapping at pre-plug. The
3480 * corresponding LMB DRCs are thus assumed to be all attachable.
3482 spapr_drc_attach(drc, dev);
3484 spapr_drc_reset(drc);
3486 addr += SPAPR_MEMORY_BLOCK_SIZE;
3488 /* send hotplug notification to the
3489 * guest only in case of hotplugged memory
3492 if (dedicated_hp_event_source) {
3493 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3494 addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3496 spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3498 spapr_drc_index(drc));
3500 spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3506 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3508 SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3509 PCDIMMDevice *dimm = PC_DIMM(dev);
3510 uint64_t size, addr;
3512 bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3514 size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3516 pc_dimm_plug(dimm, MACHINE(ms));
3519 addr = object_property_get_uint(OBJECT(dimm),
3520 PC_DIMM_ADDR_PROP, &error_abort);
3521 spapr_add_lmbs(dev, addr, size,
3522 spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3524 slot = object_property_get_int(OBJECT(dimm),
3525 PC_DIMM_SLOT_PROP, &error_abort);
3526 /* We should have valid slot number at this point */
3527 g_assert(slot >= 0);
3528 spapr_add_nvdimm(dev, slot);
3532 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3535 const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3536 SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3537 bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3538 PCDIMMDevice *dimm = PC_DIMM(dev);
3539 Error *local_err = NULL;
3544 if (!smc->dr_lmb_enabled) {
3545 error_setg(errp, "Memory hotplug not supported for this machine");
3549 size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3551 error_propagate(errp, local_err);
3556 if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3559 } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3560 error_setg(errp, "Hotplugged memory size must be a multiple of "
3561 "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3565 memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3567 pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3568 if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3572 pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3575 struct SpaprDimmState {
3578 QTAILQ_ENTRY(SpaprDimmState) next;
3581 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3584 SpaprDimmState *dimm_state = NULL;
3586 QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3587 if (dimm_state->dimm == dimm) {
3594 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3598 SpaprDimmState *ds = NULL;
3601 * If this request is for a DIMM whose removal had failed earlier
3602 * (due to guest's refusal to remove the LMBs), we would have this
3603 * dimm already in the pending_dimm_unplugs list. In that
3604 * case don't add again.
3606 ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3608 ds = g_new0(SpaprDimmState, 1);
3609 ds->nr_lmbs = nr_lmbs;
3611 QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3616 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3617 SpaprDimmState *dimm_state)
3619 QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3623 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3627 uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3629 uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3630 uint32_t avail_lmbs = 0;
3631 uint64_t addr_start, addr;
3634 addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3638 for (i = 0; i < nr_lmbs; i++) {
3639 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3640 addr / SPAPR_MEMORY_BLOCK_SIZE);
3645 addr += SPAPR_MEMORY_BLOCK_SIZE;
3648 return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3651 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3657 uint64_t size, addr_start, addr;
3658 g_autofree char *qapi_error = NULL;
3665 dimm = PC_DIMM(dev);
3666 ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3669 * 'ds == NULL' would mean that the DIMM doesn't have a pending
3670 * unplug state, but one of its DRC is marked as unplug_requested.
3671 * This is bad and weird enough to g_assert() out.
3675 spapr_pending_dimm_unplugs_remove(spapr, ds);
3677 size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3678 nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3680 addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3684 for (i = 0; i < nr_lmbs; i++) {
3685 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3686 addr / SPAPR_MEMORY_BLOCK_SIZE);
3689 drc->unplug_requested = false;
3690 addr += SPAPR_MEMORY_BLOCK_SIZE;
3694 * Tell QAPI that something happened and the memory
3695 * hotunplug wasn't successful. Keep sending
3696 * MEM_UNPLUG_ERROR even while sending
3697 * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3698 * MEM_UNPLUG_ERROR is due.
3700 qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3701 "for device %s", dev->id);
3703 qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3705 qapi_event_send_device_unplug_guest_error(!!dev->id, dev->id,
3706 dev->canonical_path);
3709 /* Callback to be called during DRC release. */
3710 void spapr_lmb_release(DeviceState *dev)
3712 HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3713 SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3714 SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3716 /* This information will get lost if a migration occurs
3717 * during the unplug process. In this case recover it. */
3719 ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3721 /* The DRC being examined by the caller at least must be counted */
3722 g_assert(ds->nr_lmbs);
3725 if (--ds->nr_lmbs) {
3730 * Now that all the LMBs have been removed by the guest, call the
3731 * unplug handler chain. This can never fail.
3733 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3734 object_unparent(OBJECT(dev));
3737 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3739 SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3740 SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3742 /* We really shouldn't get this far without anything to unplug */
3745 pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3746 qdev_unrealize(dev);
3747 spapr_pending_dimm_unplugs_remove(spapr, ds);
3750 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3751 DeviceState *dev, Error **errp)
3753 SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3754 PCDIMMDevice *dimm = PC_DIMM(dev);
3756 uint64_t size, addr_start, addr;
3760 if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3761 error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3765 size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3766 nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3768 addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3772 * An existing pending dimm state for this DIMM means that there is an
3773 * unplug operation in progress, waiting for the spapr_lmb_release
3774 * callback to complete the job (BQL can't cover that far). In this case,
3775 * bail out to avoid detaching DRCs that were already released.
3777 if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3778 error_setg(errp, "Memory unplug already in progress for device %s",
3783 spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3786 for (i = 0; i < nr_lmbs; i++) {
3787 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3788 addr / SPAPR_MEMORY_BLOCK_SIZE);
3791 spapr_drc_unplug_request(drc);
3792 addr += SPAPR_MEMORY_BLOCK_SIZE;
3795 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3796 addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3797 spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3798 nr_lmbs, spapr_drc_index(drc));
3801 /* Callback to be called during DRC release. */
3802 void spapr_core_release(DeviceState *dev)
3804 HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3806 /* Call the unplug handler chain. This can never fail. */
3807 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3808 object_unparent(OBJECT(dev));
3811 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3813 MachineState *ms = MACHINE(hotplug_dev);
3814 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3815 CPUCore *cc = CPU_CORE(dev);
3816 CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3818 if (smc->pre_2_10_has_unused_icps) {
3819 SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3822 for (i = 0; i < cc->nr_threads; i++) {
3823 CPUState *cs = CPU(sc->threads[i]);
3825 pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3830 core_slot->cpu = NULL;
3831 qdev_unrealize(dev);
3835 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3838 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3841 CPUCore *cc = CPU_CORE(dev);
3843 if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3844 error_setg(errp, "Unable to find CPU core with core-id: %d",
3849 error_setg(errp, "Boot CPU core may not be unplugged");
3853 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3854 spapr_vcpu_id(spapr, cc->core_id));
3857 if (!spapr_drc_unplug_requested(drc)) {
3858 spapr_drc_unplug_request(drc);
3862 * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3863 * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3864 * pulses removing the same CPU. Otherwise, in an failed hotunplug
3865 * attempt (e.g. the kernel will refuse to remove the last online
3866 * CPU), we will never attempt it again because unplug_requested
3867 * will still be 'true' in that case.
3869 spapr_hotplug_req_remove_by_index(drc);
3872 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3873 void *fdt, int *fdt_start_offset, Error **errp)
3875 SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3876 CPUState *cs = CPU(core->threads[0]);
3877 PowerPCCPU *cpu = POWERPC_CPU(cs);
3878 DeviceClass *dc = DEVICE_GET_CLASS(cs);
3879 int id = spapr_get_vcpu_id(cpu);
3880 g_autofree char *nodename = NULL;
3883 nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3884 offset = fdt_add_subnode(fdt, 0, nodename);
3886 spapr_dt_cpu(cs, fdt, offset, spapr);
3889 * spapr_dt_cpu() does not fill the 'name' property in the
3890 * CPU node. The function is called during boot process, before
3891 * and after CAS, and overwriting the 'name' property written
3892 * by SLOF is not allowed.
3894 * Write it manually after spapr_dt_cpu(). This makes the hotplug
3895 * CPUs more compatible with the coldplugged ones, which have
3896 * the 'name' property. Linux Kernel also relies on this
3897 * property to identify CPU nodes.
3899 _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3901 *fdt_start_offset = offset;
3905 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3907 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3908 MachineClass *mc = MACHINE_GET_CLASS(spapr);
3909 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3910 SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3911 CPUCore *cc = CPU_CORE(dev);
3914 CPUArchId *core_slot;
3916 bool hotplugged = spapr_drc_hotplugged(dev);
3919 core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3920 g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3922 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3923 spapr_vcpu_id(spapr, cc->core_id));
3925 g_assert(drc || !mc->has_hotpluggable_cpus);
3929 * spapr_core_pre_plug() already buys us this is a brand new
3930 * core being plugged into a free slot. Nothing should already
3931 * be attached to the corresponding DRC.
3933 spapr_drc_attach(drc, dev);
3937 * Send hotplug notification interrupt to the guest only
3938 * in case of hotplugged CPUs.
3940 spapr_hotplug_req_add_by_index(drc);
3942 spapr_drc_reset(drc);
3946 core_slot->cpu = OBJECT(dev);
3949 * Set compatibility mode to match the boot CPU, which was either set
3950 * by the machine reset code or by CAS. This really shouldn't fail at
3954 for (i = 0; i < cc->nr_threads; i++) {
3955 ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3960 if (smc->pre_2_10_has_unused_icps) {
3961 for (i = 0; i < cc->nr_threads; i++) {
3962 cs = CPU(core->threads[i]);
3963 pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3968 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3971 MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3972 MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3973 CPUCore *cc = CPU_CORE(dev);
3974 const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3975 const char *type = object_get_typename(OBJECT(dev));
3976 CPUArchId *core_slot;
3978 unsigned int smp_threads = machine->smp.threads;
3980 if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3981 error_setg(errp, "CPU hotplug not supported for this machine");
3985 if (strcmp(base_core_type, type)) {
3986 error_setg(errp, "CPU core type should be %s", base_core_type);
3990 if (cc->core_id % smp_threads) {
3991 error_setg(errp, "invalid core id %d", cc->core_id);
3996 * In general we should have homogeneous threads-per-core, but old
3997 * (pre hotplug support) machine types allow the last core to have
3998 * reduced threads as a compatibility hack for when we allowed
3999 * total vcpus not a multiple of threads-per-core.
4001 if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
4002 error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
4007 core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4009 error_setg(errp, "core id %d out of range", cc->core_id);
4013 if (core_slot->cpu) {
4014 error_setg(errp, "core %d already populated", cc->core_id);
4018 numa_cpu_pre_plug(core_slot, dev, errp);
4021 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
4022 void *fdt, int *fdt_start_offset, Error **errp)
4024 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4027 intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4028 if (intc_phandle <= 0) {
4032 if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4033 error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4037 /* generally SLOF creates these, for hotplug it's up to QEMU */
4038 _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4043 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4046 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4047 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4048 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4049 const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4052 if (dev->hotplugged && !smc->dr_phb_enabled) {
4053 error_setg(errp, "PHB hotplug not supported for this machine");
4057 if (sphb->index == (uint32_t)-1) {
4058 error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4062 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4063 if (drc && drc->dev) {
4064 error_setg(errp, "PHB %d already attached", sphb->index);
4069 * This will check that sphb->index doesn't exceed the maximum number of
4070 * PHBs for the current machine type.
4073 smc->phb_placement(spapr, sphb->index,
4074 &sphb->buid, &sphb->io_win_addr,
4075 &sphb->mem_win_addr, &sphb->mem64_win_addr,
4076 windows_supported, sphb->dma_liobn,
4077 &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4081 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4083 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4084 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4085 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4087 bool hotplugged = spapr_drc_hotplugged(dev);
4089 if (!smc->dr_phb_enabled) {
4093 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4094 /* hotplug hooks should check it's enabled before getting this far */
4097 /* spapr_phb_pre_plug() already checked the DRC is attachable */
4098 spapr_drc_attach(drc, dev);
4101 spapr_hotplug_req_add_by_index(drc);
4103 spapr_drc_reset(drc);
4107 void spapr_phb_release(DeviceState *dev)
4109 HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4111 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4112 object_unparent(OBJECT(dev));
4115 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4117 qdev_unrealize(dev);
4120 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4121 DeviceState *dev, Error **errp)
4123 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4126 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4129 if (!spapr_drc_unplug_requested(drc)) {
4130 spapr_drc_unplug_request(drc);
4131 spapr_hotplug_req_remove_by_index(drc);
4134 "PCI Host Bridge unplug already in progress for device %s",
4140 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4143 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4145 if (spapr->tpm_proxy != NULL) {
4146 error_setg(errp, "Only one TPM proxy can be specified for this machine");
4153 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4155 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4156 SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4158 /* Already checked in spapr_tpm_proxy_pre_plug() */
4159 g_assert(spapr->tpm_proxy == NULL);
4161 spapr->tpm_proxy = tpm_proxy;
4164 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4166 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4168 qdev_unrealize(dev);
4169 object_unparent(OBJECT(dev));
4170 spapr->tpm_proxy = NULL;
4173 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4174 DeviceState *dev, Error **errp)
4176 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4177 spapr_memory_plug(hotplug_dev, dev);
4178 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4179 spapr_core_plug(hotplug_dev, dev);
4180 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4181 spapr_phb_plug(hotplug_dev, dev);
4182 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4183 spapr_tpm_proxy_plug(hotplug_dev, dev);
4187 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4188 DeviceState *dev, Error **errp)
4190 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4191 spapr_memory_unplug(hotplug_dev, dev);
4192 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4193 spapr_core_unplug(hotplug_dev, dev);
4194 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4195 spapr_phb_unplug(hotplug_dev, dev);
4196 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4197 spapr_tpm_proxy_unplug(hotplug_dev, dev);
4201 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4203 return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4205 * CAS will process all pending unplug requests.
4207 * HACK: a guest could theoretically have cleared all bits in OV5,
4208 * but none of the guests we care for do.
4210 spapr_ovec_empty(spapr->ov5_cas);
4213 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4214 DeviceState *dev, Error **errp)
4216 SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4217 MachineClass *mc = MACHINE_GET_CLASS(sms);
4218 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4220 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4221 if (spapr_memory_hot_unplug_supported(sms)) {
4222 spapr_memory_unplug_request(hotplug_dev, dev, errp);
4224 error_setg(errp, "Memory hot unplug not supported for this guest");
4226 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4227 if (!mc->has_hotpluggable_cpus) {
4228 error_setg(errp, "CPU hot unplug not supported on this machine");
4231 spapr_core_unplug_request(hotplug_dev, dev, errp);
4232 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4233 if (!smc->dr_phb_enabled) {
4234 error_setg(errp, "PHB hot unplug not supported on this machine");
4237 spapr_phb_unplug_request(hotplug_dev, dev, errp);
4238 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4239 spapr_tpm_proxy_unplug(hotplug_dev, dev);
4243 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4244 DeviceState *dev, Error **errp)
4246 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4247 spapr_memory_pre_plug(hotplug_dev, dev, errp);
4248 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4249 spapr_core_pre_plug(hotplug_dev, dev, errp);
4250 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4251 spapr_phb_pre_plug(hotplug_dev, dev, errp);
4252 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4253 spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4257 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4260 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4261 object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4262 object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4263 object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4264 return HOTPLUG_HANDLER(machine);
4266 if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4267 PCIDevice *pcidev = PCI_DEVICE(dev);
4268 PCIBus *root = pci_device_root_bus(pcidev);
4269 SpaprPhbState *phb =
4270 (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4271 TYPE_SPAPR_PCI_HOST_BRIDGE);
4274 return HOTPLUG_HANDLER(phb);
4280 static CpuInstanceProperties
4281 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4283 CPUArchId *core_slot;
4284 MachineClass *mc = MACHINE_GET_CLASS(machine);
4286 /* make sure possible_cpu are intialized */
4287 mc->possible_cpu_arch_ids(machine);
4288 /* get CPU core slot containing thread that matches cpu_index */
4289 core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4291 return core_slot->props;
4294 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4296 return idx / ms->smp.cores % ms->numa_state->num_nodes;
4299 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4302 unsigned int smp_threads = machine->smp.threads;
4303 unsigned int smp_cpus = machine->smp.cpus;
4304 const char *core_type;
4305 int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4306 MachineClass *mc = MACHINE_GET_CLASS(machine);
4308 if (!mc->has_hotpluggable_cpus) {
4309 spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4311 if (machine->possible_cpus) {
4312 assert(machine->possible_cpus->len == spapr_max_cores);
4313 return machine->possible_cpus;
4316 core_type = spapr_get_cpu_core_type(machine->cpu_type);
4318 error_report("Unable to find sPAPR CPU Core definition");
4322 machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4323 sizeof(CPUArchId) * spapr_max_cores);
4324 machine->possible_cpus->len = spapr_max_cores;
4325 for (i = 0; i < machine->possible_cpus->len; i++) {
4326 int core_id = i * smp_threads;
4328 machine->possible_cpus->cpus[i].type = core_type;
4329 machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4330 machine->possible_cpus->cpus[i].arch_id = core_id;
4331 machine->possible_cpus->cpus[i].props.has_core_id = true;
4332 machine->possible_cpus->cpus[i].props.core_id = core_id;
4334 return machine->possible_cpus;
4337 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4338 uint64_t *buid, hwaddr *pio,
4339 hwaddr *mmio32, hwaddr *mmio64,
4340 unsigned n_dma, uint32_t *liobns,
4341 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4344 * New-style PHB window placement.
4346 * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4347 * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4350 * Some guest kernels can't work with MMIO windows above 1<<46
4351 * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4353 * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4354 * PHB stacked together. (32TiB+2GiB)..(32TiB+64GiB) contains the
4355 * 2GiB 32-bit MMIO windows for each PHB. Then 33..64TiB has the
4356 * 1TiB 64-bit MMIO windows for each PHB.
4358 const uint64_t base_buid = 0x800000020000000ULL;
4361 /* Sanity check natural alignments */
4362 QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4363 QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4364 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4365 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4366 /* Sanity check bounds */
4367 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4368 SPAPR_PCI_MEM32_WIN_SIZE);
4369 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4370 SPAPR_PCI_MEM64_WIN_SIZE);
4372 if (index >= SPAPR_MAX_PHBS) {
4373 error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4374 SPAPR_MAX_PHBS - 1);
4378 *buid = base_buid + index;
4379 for (i = 0; i < n_dma; ++i) {
4380 liobns[i] = SPAPR_PCI_LIOBN(index, i);
4383 *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4384 *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4385 *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4387 *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4388 *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4392 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4394 SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4396 return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4399 static void spapr_ics_resend(XICSFabric *dev)
4401 SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4403 ics_resend(spapr->ics);
4406 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4408 PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4410 return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4413 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4416 SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4418 spapr_irq_print_info(spapr, mon);
4419 monitor_printf(mon, "irqchip: %s\n",
4420 kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4424 * This is a XIVE only operation
4426 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4427 uint8_t nvt_blk, uint32_t nvt_idx,
4428 bool cam_ignore, uint8_t priority,
4429 uint32_t logic_serv, XiveTCTXMatch *match)
4431 SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4432 XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4433 XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4436 count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4437 priority, logic_serv, match);
4443 * When we implement the save and restore of the thread interrupt
4444 * contexts in the enter/exit CPU handlers of the machine and the
4445 * escalations in QEMU, we should be able to handle non dispatched
4448 * Until this is done, the sPAPR machine should find at least one
4449 * matching context always.
4452 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4459 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4461 return cpu->vcpu_id;
4464 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4466 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4467 MachineState *ms = MACHINE(spapr);
4470 vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4472 if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4473 error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4474 error_append_hint(errp, "Adjust the number of cpus to %d "
4475 "or try to raise the number of threads per core\n",
4476 vcpu_id * ms->smp.threads / spapr->vsmt);
4480 cpu->vcpu_id = vcpu_id;
4484 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4489 PowerPCCPU *cpu = POWERPC_CPU(cs);
4491 if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4499 static bool spapr_cpu_in_nested(PowerPCCPU *cpu)
4501 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4503 return spapr_cpu->in_nested;
4506 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4508 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4510 /* These are only called by TCG, KVM maintains dispatch state */
4512 spapr_cpu->prod = false;
4513 if (spapr_cpu->vpa_addr) {
4514 CPUState *cs = CPU(cpu);
4517 dispatch = ldl_be_phys(cs->as,
4518 spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4520 if ((dispatch & 1) != 0) {
4521 qemu_log_mask(LOG_GUEST_ERROR,
4522 "VPA: incorrect dispatch counter value for "
4523 "dispatched partition %u, correcting.\n", dispatch);
4527 spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4531 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4533 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4535 if (spapr_cpu->vpa_addr) {
4536 CPUState *cs = CPU(cpu);
4539 dispatch = ldl_be_phys(cs->as,
4540 spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4542 if ((dispatch & 1) != 1) {
4543 qemu_log_mask(LOG_GUEST_ERROR,
4544 "VPA: incorrect dispatch counter value for "
4545 "preempted partition %u, correcting.\n", dispatch);
4549 spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4553 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4555 MachineClass *mc = MACHINE_CLASS(oc);
4556 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4557 FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4558 NMIClass *nc = NMI_CLASS(oc);
4559 HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4560 PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4561 XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4562 InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4563 XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4564 VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4566 mc->desc = "pSeries Logical Partition (PAPR compliant)";
4567 mc->ignore_boot_device_suffixes = true;
4570 * We set up the default / latest behaviour here. The class_init
4571 * functions for the specific versioned machine types can override
4572 * these details for backwards compatibility
4574 mc->init = spapr_machine_init;
4575 mc->reset = spapr_machine_reset;
4576 mc->block_default_type = IF_SCSI;
4579 * Setting max_cpus to INT32_MAX. Both KVM and TCG max_cpus values
4580 * should be limited by the host capability instead of hardcoded.
4581 * max_cpus for KVM guests will be checked in kvm_init(), and TCG
4582 * guests are welcome to have as many CPUs as the host are capable
4585 mc->max_cpus = INT32_MAX;
4587 mc->no_parallel = 1;
4588 mc->default_boot_order = "";
4589 mc->default_ram_size = 512 * MiB;
4590 mc->default_ram_id = "ppc_spapr.ram";
4591 mc->default_display = "std";
4592 mc->kvm_type = spapr_kvm_type;
4593 machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4594 mc->pci_allow_0_address = true;
4595 assert(!mc->get_hotplug_handler);
4596 mc->get_hotplug_handler = spapr_get_hotplug_handler;
4597 hc->pre_plug = spapr_machine_device_pre_plug;
4598 hc->plug = spapr_machine_device_plug;
4599 mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4600 mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4601 mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4602 hc->unplug_request = spapr_machine_device_unplug_request;
4603 hc->unplug = spapr_machine_device_unplug;
4605 smc->dr_lmb_enabled = true;
4606 smc->update_dt_enabled = true;
4607 mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4608 mc->has_hotpluggable_cpus = true;
4609 mc->nvdimm_supported = true;
4610 smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4611 fwc->get_dev_path = spapr_get_fw_dev_path;
4612 nc->nmi_monitor_handler = spapr_nmi;
4613 smc->phb_placement = spapr_phb_placement;
4614 vhc->cpu_in_nested = spapr_cpu_in_nested;
4615 vhc->deliver_hv_excp = spapr_exit_nested;
4616 vhc->hypercall = emulate_spapr_hypercall;
4617 vhc->hpt_mask = spapr_hpt_mask;
4618 vhc->map_hptes = spapr_map_hptes;
4619 vhc->unmap_hptes = spapr_unmap_hptes;
4620 vhc->hpte_set_c = spapr_hpte_set_c;
4621 vhc->hpte_set_r = spapr_hpte_set_r;
4622 vhc->get_pate = spapr_get_pate;
4623 vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4624 vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4625 vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4626 xic->ics_get = spapr_ics_get;
4627 xic->ics_resend = spapr_ics_resend;
4628 xic->icp_get = spapr_icp_get;
4629 ispc->print_info = spapr_pic_print_info;
4630 /* Force NUMA node memory size to be a multiple of
4631 * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4632 * in which LMBs are represented and hot-added
4634 mc->numa_mem_align_shift = 28;
4635 mc->auto_enable_numa = true;
4637 smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4638 smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4639 smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4640 smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4641 smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4642 smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4643 smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4644 smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4645 smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4646 smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4647 smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4648 smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4649 spapr_caps_add_properties(smc);
4650 smc->irq = &spapr_irq_dual;
4651 smc->dr_phb_enabled = true;
4652 smc->linux_pci_probe = true;
4653 smc->smp_threads_vsmt = true;
4654 smc->nr_xirqs = SPAPR_NR_XIRQS;
4655 xfc->match_nvt = spapr_match_nvt;
4656 vmc->client_architecture_support = spapr_vof_client_architecture_support;
4657 vmc->quiesce = spapr_vof_quiesce;
4658 vmc->setprop = spapr_vof_setprop;
4661 static const TypeInfo spapr_machine_info = {
4662 .name = TYPE_SPAPR_MACHINE,
4663 .parent = TYPE_MACHINE,
4665 .instance_size = sizeof(SpaprMachineState),
4666 .instance_init = spapr_instance_init,
4667 .instance_finalize = spapr_machine_finalizefn,
4668 .class_size = sizeof(SpaprMachineClass),
4669 .class_init = spapr_machine_class_init,
4670 .interfaces = (InterfaceInfo[]) {
4671 { TYPE_FW_PATH_PROVIDER },
4673 { TYPE_HOTPLUG_HANDLER },
4674 { TYPE_PPC_VIRTUAL_HYPERVISOR },
4675 { TYPE_XICS_FABRIC },
4676 { TYPE_INTERRUPT_STATS_PROVIDER },
4677 { TYPE_XIVE_FABRIC },
4678 { TYPE_VOF_MACHINE_IF },
4683 static void spapr_machine_latest_class_options(MachineClass *mc)
4685 mc->alias = "pseries";
4686 mc->is_default = true;
4689 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest) \
4690 static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4693 MachineClass *mc = MACHINE_CLASS(oc); \
4694 spapr_machine_##suffix##_class_options(mc); \
4696 spapr_machine_latest_class_options(mc); \
4699 static const TypeInfo spapr_machine_##suffix##_info = { \
4700 .name = MACHINE_TYPE_NAME("pseries-" verstr), \
4701 .parent = TYPE_SPAPR_MACHINE, \
4702 .class_init = spapr_machine_##suffix##_class_init, \
4704 static void spapr_machine_register_##suffix(void) \
4706 type_register(&spapr_machine_##suffix##_info); \
4708 type_init(spapr_machine_register_##suffix)
4713 static void spapr_machine_7_1_class_options(MachineClass *mc)
4715 /* Defaults for the latest behaviour inherited from the base class */
4718 DEFINE_SPAPR_MACHINE(7_1, "7.1", true);
4723 static void spapr_machine_7_0_class_options(MachineClass *mc)
4725 spapr_machine_7_1_class_options(mc);
4726 compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
4729 DEFINE_SPAPR_MACHINE(7_0, "7.0", false);
4734 static void spapr_machine_6_2_class_options(MachineClass *mc)
4736 spapr_machine_7_0_class_options(mc);
4737 compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4740 DEFINE_SPAPR_MACHINE(6_2, "6.2", false);
4745 static void spapr_machine_6_1_class_options(MachineClass *mc)
4747 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4749 spapr_machine_6_2_class_options(mc);
4750 compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4751 smc->pre_6_2_numa_affinity = true;
4752 mc->smp_props.prefer_sockets = true;
4755 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4760 static void spapr_machine_6_0_class_options(MachineClass *mc)
4762 spapr_machine_6_1_class_options(mc);
4763 compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4766 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4771 static void spapr_machine_5_2_class_options(MachineClass *mc)
4773 spapr_machine_6_0_class_options(mc);
4774 compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4777 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4782 static void spapr_machine_5_1_class_options(MachineClass *mc)
4784 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4786 spapr_machine_5_2_class_options(mc);
4787 compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4788 smc->pre_5_2_numa_associativity = true;
4791 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4796 static void spapr_machine_5_0_class_options(MachineClass *mc)
4798 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4799 static GlobalProperty compat[] = {
4800 { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4803 spapr_machine_5_1_class_options(mc);
4804 compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4805 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4806 mc->numa_mem_supported = true;
4807 smc->pre_5_1_assoc_refpoints = true;
4810 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4815 static void spapr_machine_4_2_class_options(MachineClass *mc)
4817 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4819 spapr_machine_5_0_class_options(mc);
4820 compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4821 smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4822 smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4823 smc->rma_limit = 16 * GiB;
4824 mc->nvdimm_supported = false;
4827 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4832 static void spapr_machine_4_1_class_options(MachineClass *mc)
4834 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4835 static GlobalProperty compat[] = {
4836 /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4837 { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4840 spapr_machine_4_2_class_options(mc);
4841 smc->linux_pci_probe = false;
4842 smc->smp_threads_vsmt = false;
4843 compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4844 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4847 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4852 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4853 uint64_t *buid, hwaddr *pio,
4854 hwaddr *mmio32, hwaddr *mmio64,
4855 unsigned n_dma, uint32_t *liobns,
4856 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4858 if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4859 liobns, nv2gpa, nv2atsd, errp)) {
4867 static void spapr_machine_4_0_class_options(MachineClass *mc)
4869 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4871 spapr_machine_4_1_class_options(mc);
4872 compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4873 smc->phb_placement = phb_placement_4_0;
4874 smc->irq = &spapr_irq_xics;
4875 smc->pre_4_1_migration = true;
4878 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4883 static void spapr_machine_3_1_class_options(MachineClass *mc)
4885 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4887 spapr_machine_4_0_class_options(mc);
4888 compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4890 mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4891 smc->update_dt_enabled = false;
4892 smc->dr_phb_enabled = false;
4893 smc->broken_host_serial_model = true;
4894 smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4895 smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4896 smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4897 smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4900 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4906 static void spapr_machine_3_0_class_options(MachineClass *mc)
4908 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4910 spapr_machine_3_1_class_options(mc);
4911 compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4913 smc->legacy_irq_allocation = true;
4914 smc->nr_xirqs = 0x400;
4915 smc->irq = &spapr_irq_xics_legacy;
4918 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4923 static void spapr_machine_2_12_class_options(MachineClass *mc)
4925 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4926 static GlobalProperty compat[] = {
4927 { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4928 { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4931 spapr_machine_3_0_class_options(mc);
4932 compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4933 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4935 /* We depend on kvm_enabled() to choose a default value for the
4936 * hpt-max-page-size capability. Of course we can't do it here
4937 * because this is too early and the HW accelerator isn't initialzed
4938 * yet. Postpone this to machine init (see default_caps_with_cpu()).
4940 smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4943 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4945 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4947 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4949 spapr_machine_2_12_class_options(mc);
4950 smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4951 smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4952 smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4955 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4961 static void spapr_machine_2_11_class_options(MachineClass *mc)
4963 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4965 spapr_machine_2_12_class_options(mc);
4966 smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4967 compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4970 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4976 static void spapr_machine_2_10_class_options(MachineClass *mc)
4978 spapr_machine_2_11_class_options(mc);
4979 compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4982 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4988 static void spapr_machine_2_9_class_options(MachineClass *mc)
4990 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4991 static GlobalProperty compat[] = {
4992 { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4995 spapr_machine_2_10_class_options(mc);
4996 compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4997 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4998 smc->pre_2_10_has_unused_icps = true;
4999 smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
5002 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
5008 static void spapr_machine_2_8_class_options(MachineClass *mc)
5010 static GlobalProperty compat[] = {
5011 { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
5014 spapr_machine_2_9_class_options(mc);
5015 compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
5016 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5017 mc->numa_mem_align_shift = 23;
5020 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
5026 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
5027 uint64_t *buid, hwaddr *pio,
5028 hwaddr *mmio32, hwaddr *mmio64,
5029 unsigned n_dma, uint32_t *liobns,
5030 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
5032 /* Legacy PHB placement for pseries-2.7 and earlier machine types */
5033 const uint64_t base_buid = 0x800000020000000ULL;
5034 const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
5035 const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
5036 const hwaddr pio_offset = 0x80000000; /* 2 GiB */
5037 const uint32_t max_index = 255;
5038 const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
5040 uint64_t ram_top = MACHINE(spapr)->ram_size;
5041 hwaddr phb0_base, phb_base;
5044 /* Do we have device memory? */
5045 if (MACHINE(spapr)->maxram_size > ram_top) {
5046 /* Can't just use maxram_size, because there may be an
5047 * alignment gap between normal and device memory regions
5049 ram_top = MACHINE(spapr)->device_memory->base +
5050 memory_region_size(&MACHINE(spapr)->device_memory->mr);
5053 phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5055 if (index > max_index) {
5056 error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5061 *buid = base_buid + index;
5062 for (i = 0; i < n_dma; ++i) {
5063 liobns[i] = SPAPR_PCI_LIOBN(index, i);
5066 phb_base = phb0_base + index * phb_spacing;
5067 *pio = phb_base + pio_offset;
5068 *mmio32 = phb_base + mmio_offset;
5070 * We don't set the 64-bit MMIO window, relying on the PHB's
5071 * fallback behaviour of automatically splitting a large "32-bit"
5072 * window into contiguous 32-bit and 64-bit windows
5080 static void spapr_machine_2_7_class_options(MachineClass *mc)
5082 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5083 static GlobalProperty compat[] = {
5084 { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5085 { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5086 { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5087 { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5090 spapr_machine_2_8_class_options(mc);
5091 mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5092 mc->default_machine_opts = "modern-hotplug-events=off";
5093 compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5094 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5095 smc->phb_placement = phb_placement_2_7;
5098 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5104 static void spapr_machine_2_6_class_options(MachineClass *mc)
5106 static GlobalProperty compat[] = {
5107 { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5110 spapr_machine_2_7_class_options(mc);
5111 mc->has_hotpluggable_cpus = false;
5112 compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5113 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5116 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5122 static void spapr_machine_2_5_class_options(MachineClass *mc)
5124 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5125 static GlobalProperty compat[] = {
5126 { "spapr-vlan", "use-rx-buffer-pools", "off" },
5129 spapr_machine_2_6_class_options(mc);
5130 smc->use_ohci_by_default = true;
5131 compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5132 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5135 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5141 static void spapr_machine_2_4_class_options(MachineClass *mc)
5143 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5145 spapr_machine_2_5_class_options(mc);
5146 smc->dr_lmb_enabled = false;
5147 compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5150 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5156 static void spapr_machine_2_3_class_options(MachineClass *mc)
5158 static GlobalProperty compat[] = {
5159 { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5161 spapr_machine_2_4_class_options(mc);
5162 compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5163 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5165 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5171 static void spapr_machine_2_2_class_options(MachineClass *mc)
5173 static GlobalProperty compat[] = {
5174 { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5177 spapr_machine_2_3_class_options(mc);
5178 compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5179 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5180 mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5182 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5188 static void spapr_machine_2_1_class_options(MachineClass *mc)
5190 spapr_machine_2_2_class_options(mc);
5191 compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5193 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5195 static void spapr_machine_register_types(void)
5197 type_register_static(&spapr_machine_info);
5200 type_init(spapr_machine_register_types)