2 * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
4 * Copyright (c) 2004-2007 Fabrice Bellard
5 * Copyright (c) 2007 Jocelyn Mayer
6 * Copyright (c) 2010 David Gibson, IBM Corporation.
8 * Permission is hereby granted, free of charge, to any person obtaining a copy
9 * of this software and associated documentation files (the "Software"), to deal
10 * in the Software without restriction, including without limitation the rights
11 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12 * copies of the Software, and to permit persons to whom the Software is
13 * furnished to do so, subject to the following conditions:
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qemu/datadir.h"
30 #include "qapi/error.h"
31 #include "qapi/qapi-events-machine.h"
32 #include "qapi/visitor.h"
33 #include "sysemu/sysemu.h"
34 #include "sysemu/hostmem.h"
35 #include "sysemu/numa.h"
36 #include "sysemu/qtest.h"
37 #include "sysemu/reset.h"
38 #include "sysemu/runstate.h"
40 #include "hw/fw-path-provider.h"
43 #include "sysemu/device_tree.h"
44 #include "sysemu/cpus.h"
45 #include "sysemu/hw_accel.h"
47 #include "migration/misc.h"
48 #include "migration/qemu-file-types.h"
49 #include "migration/global_state.h"
50 #include "migration/register.h"
51 #include "migration/blocker.h"
52 #include "mmu-hash64.h"
53 #include "mmu-book3s-v3.h"
54 #include "cpu-models.h"
55 #include "hw/core/cpu.h"
57 #include "hw/ppc/ppc.h"
58 #include "hw/loader.h"
60 #include "hw/ppc/fdt.h"
61 #include "hw/ppc/spapr.h"
62 #include "hw/ppc/spapr_vio.h"
63 #include "hw/qdev-properties.h"
64 #include "hw/pci-host/spapr.h"
65 #include "hw/pci/msi.h"
67 #include "hw/pci/pci.h"
68 #include "hw/scsi/scsi.h"
69 #include "hw/virtio/virtio-scsi.h"
70 #include "hw/virtio/vhost-scsi-common.h"
72 #include "exec/ram_addr.h"
74 #include "qemu/config-file.h"
75 #include "qemu/error-report.h"
78 #include "hw/intc/intc.h"
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 #include "hw/ppc/spapr_nvdimm.h"
84 #include "hw/ppc/spapr_numa.h"
85 #include "hw/ppc/pef.h"
87 #include "monitor/monitor.h"
91 /* SLOF memory layout:
93 * SLOF raw image loaded at 0, copies its romfs right below the flat
94 * device-tree, then position SLOF itself 31M below that
96 * So we set FW_OVERHEAD to 40MB which should account for all of that
99 * We load our kernel at 4M, leaving space for SLOF initial image
101 #define FDT_MAX_ADDR 0x80000000 /* FDT must stay below that */
102 #define FW_MAX_SIZE 0x400000
103 #define FW_FILE_NAME "slof.bin"
104 #define FW_OVERHEAD 0x2800000
105 #define KERNEL_LOAD_ADDR FW_MAX_SIZE
107 #define MIN_RMA_SLOF (128 * MiB)
109 #define PHANDLE_INTC 0x00001111
111 /* These two functions implement the VCPU id numbering: one to compute them
112 * all and one to identify thread 0 of a VCORE. Any change to the first one
113 * is likely to have an impact on the second one, so let's keep them close.
115 static int spapr_vcpu_id(SpaprMachineState
*spapr
, int cpu_index
)
117 MachineState
*ms
= MACHINE(spapr
);
118 unsigned int smp_threads
= ms
->smp
.threads
;
122 (cpu_index
/ smp_threads
) * spapr
->vsmt
+ cpu_index
% smp_threads
;
124 static bool spapr_is_thread0_in_vcore(SpaprMachineState
*spapr
,
128 return spapr_get_vcpu_id(cpu
) % spapr
->vsmt
== 0;
131 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque
)
133 /* Dummy entries correspond to unused ICPState objects in older QEMUs,
134 * and newer QEMUs don't even have them. In both cases, we don't want
135 * to send anything on the wire.
140 static const VMStateDescription pre_2_10_vmstate_dummy_icp
= {
141 .name
= "icp/server",
143 .minimum_version_id
= 1,
144 .needed
= pre_2_10_vmstate_dummy_icp_needed
,
145 .fields
= (VMStateField
[]) {
146 VMSTATE_UNUSED(4), /* uint32_t xirr */
147 VMSTATE_UNUSED(1), /* uint8_t pending_priority */
148 VMSTATE_UNUSED(1), /* uint8_t mfrr */
149 VMSTATE_END_OF_LIST()
153 static void pre_2_10_vmstate_register_dummy_icp(int i
)
155 vmstate_register(NULL
, i
, &pre_2_10_vmstate_dummy_icp
,
156 (void *)(uintptr_t) i
);
159 static void pre_2_10_vmstate_unregister_dummy_icp(int i
)
161 vmstate_unregister(NULL
, &pre_2_10_vmstate_dummy_icp
,
162 (void *)(uintptr_t) i
);
165 int spapr_max_server_number(SpaprMachineState
*spapr
)
167 MachineState
*ms
= MACHINE(spapr
);
170 return DIV_ROUND_UP(ms
->smp
.max_cpus
* spapr
->vsmt
, ms
->smp
.threads
);
173 static int spapr_fixup_cpu_smt_dt(void *fdt
, int offset
, PowerPCCPU
*cpu
,
177 uint32_t servers_prop
[smt_threads
];
178 uint32_t gservers_prop
[smt_threads
* 2];
179 int index
= spapr_get_vcpu_id(cpu
);
181 if (cpu
->compat_pvr
) {
182 ret
= fdt_setprop_cell(fdt
, offset
, "cpu-version", cpu
->compat_pvr
);
188 /* Build interrupt servers and gservers properties */
189 for (i
= 0; i
< smt_threads
; i
++) {
190 servers_prop
[i
] = cpu_to_be32(index
+ i
);
191 /* Hack, direct the group queues back to cpu 0 */
192 gservers_prop
[i
*2] = cpu_to_be32(index
+ i
);
193 gservers_prop
[i
*2 + 1] = 0;
195 ret
= fdt_setprop(fdt
, offset
, "ibm,ppc-interrupt-server#s",
196 servers_prop
, sizeof(servers_prop
));
200 ret
= fdt_setprop(fdt
, offset
, "ibm,ppc-interrupt-gserver#s",
201 gservers_prop
, sizeof(gservers_prop
));
206 static void spapr_dt_pa_features(SpaprMachineState
*spapr
,
208 void *fdt
, int offset
)
210 uint8_t pa_features_206
[] = { 6, 0,
211 0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
212 uint8_t pa_features_207
[] = { 24, 0,
213 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
214 0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
215 0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
216 0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
217 uint8_t pa_features_300
[] = { 66, 0,
218 /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
219 /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
220 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
222 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
224 0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
225 /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
226 0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
227 /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
228 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
229 /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
230 0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
231 /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
232 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
233 /* 42: PM, 44: PC RA, 46: SC vec'd */
234 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
235 /* 48: SIMD, 50: QP BFP, 52: String */
236 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
237 /* 54: DecFP, 56: DecI, 58: SHA */
238 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
239 /* 60: NM atomic, 62: RNG */
240 0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
242 uint8_t *pa_features
= NULL
;
245 if (ppc_check_compat(cpu
, CPU_POWERPC_LOGICAL_2_06
, 0, cpu
->compat_pvr
)) {
246 pa_features
= pa_features_206
;
247 pa_size
= sizeof(pa_features_206
);
249 if (ppc_check_compat(cpu
, CPU_POWERPC_LOGICAL_2_07
, 0, cpu
->compat_pvr
)) {
250 pa_features
= pa_features_207
;
251 pa_size
= sizeof(pa_features_207
);
253 if (ppc_check_compat(cpu
, CPU_POWERPC_LOGICAL_3_00
, 0, cpu
->compat_pvr
)) {
254 pa_features
= pa_features_300
;
255 pa_size
= sizeof(pa_features_300
);
261 if (ppc_hash64_has(cpu
, PPC_HASH64_CI_LARGEPAGE
)) {
263 * Note: we keep CI large pages off by default because a 64K capable
264 * guest provisioned with large pages might otherwise try to map a qemu
265 * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
266 * even if that qemu runs on a 4k host.
267 * We dd this bit back here if we are confident this is not an issue
269 pa_features
[3] |= 0x20;
271 if ((spapr_get_cap(spapr
, SPAPR_CAP_HTM
) != 0) && pa_size
> 24) {
272 pa_features
[24] |= 0x80; /* Transactional memory support */
274 if (spapr
->cas_pre_isa3_guest
&& pa_size
> 40) {
275 /* Workaround for broken kernels that attempt (guest) radix
276 * mode when they can't handle it, if they see the radix bit set
277 * in pa-features. So hide it from them. */
278 pa_features
[40 + 2] &= ~0x80; /* Radix MMU */
281 _FDT((fdt_setprop(fdt
, offset
, "ibm,pa-features", pa_features
, pa_size
)));
284 static hwaddr
spapr_node0_size(MachineState
*machine
)
286 if (machine
->numa_state
->num_nodes
) {
288 for (i
= 0; i
< machine
->numa_state
->num_nodes
; ++i
) {
289 if (machine
->numa_state
->nodes
[i
].node_mem
) {
290 return MIN(pow2floor(machine
->numa_state
->nodes
[i
].node_mem
),
295 return machine
->ram_size
;
298 static void add_str(GString
*s
, const gchar
*s1
)
300 g_string_append_len(s
, s1
, strlen(s1
) + 1);
303 static int spapr_dt_memory_node(SpaprMachineState
*spapr
, void *fdt
, int nodeid
,
304 hwaddr start
, hwaddr size
)
307 uint64_t mem_reg_property
[2];
310 mem_reg_property
[0] = cpu_to_be64(start
);
311 mem_reg_property
[1] = cpu_to_be64(size
);
313 sprintf(mem_name
, "memory@%" HWADDR_PRIx
, start
);
314 off
= fdt_add_subnode(fdt
, 0, mem_name
);
316 _FDT((fdt_setprop_string(fdt
, off
, "device_type", "memory")));
317 _FDT((fdt_setprop(fdt
, off
, "reg", mem_reg_property
,
318 sizeof(mem_reg_property
))));
319 spapr_numa_write_associativity_dt(spapr
, fdt
, off
, nodeid
);
323 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList
*list
, ram_addr_t addr
)
325 MemoryDeviceInfoList
*info
;
327 for (info
= list
; info
; info
= info
->next
) {
328 MemoryDeviceInfo
*value
= info
->value
;
330 if (value
&& value
->type
== MEMORY_DEVICE_INFO_KIND_DIMM
) {
331 PCDIMMDeviceInfo
*pcdimm_info
= value
->u
.dimm
.data
;
333 if (addr
>= pcdimm_info
->addr
&&
334 addr
< (pcdimm_info
->addr
+ pcdimm_info
->size
)) {
335 return pcdimm_info
->node
;
343 struct sPAPRDrconfCellV2
{
351 typedef struct DrconfCellQueue
{
352 struct sPAPRDrconfCellV2 cell
;
353 QSIMPLEQ_ENTRY(DrconfCellQueue
) entry
;
356 static DrconfCellQueue
*
357 spapr_get_drconf_cell(uint32_t seq_lmbs
, uint64_t base_addr
,
358 uint32_t drc_index
, uint32_t aa_index
,
361 DrconfCellQueue
*elem
;
363 elem
= g_malloc0(sizeof(*elem
));
364 elem
->cell
.seq_lmbs
= cpu_to_be32(seq_lmbs
);
365 elem
->cell
.base_addr
= cpu_to_be64(base_addr
);
366 elem
->cell
.drc_index
= cpu_to_be32(drc_index
);
367 elem
->cell
.aa_index
= cpu_to_be32(aa_index
);
368 elem
->cell
.flags
= cpu_to_be32(flags
);
373 static int spapr_dt_dynamic_memory_v2(SpaprMachineState
*spapr
, void *fdt
,
374 int offset
, MemoryDeviceInfoList
*dimms
)
376 MachineState
*machine
= MACHINE(spapr
);
377 uint8_t *int_buf
, *cur_index
;
379 uint64_t lmb_size
= SPAPR_MEMORY_BLOCK_SIZE
;
380 uint64_t addr
, cur_addr
, size
;
381 uint32_t nr_boot_lmbs
= (machine
->device_memory
->base
/ lmb_size
);
382 uint64_t mem_end
= machine
->device_memory
->base
+
383 memory_region_size(&machine
->device_memory
->mr
);
384 uint32_t node
, buf_len
, nr_entries
= 0;
386 DrconfCellQueue
*elem
, *next
;
387 MemoryDeviceInfoList
*info
;
388 QSIMPLEQ_HEAD(, DrconfCellQueue
) drconf_queue
389 = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue
);
391 /* Entry to cover RAM and the gap area */
392 elem
= spapr_get_drconf_cell(nr_boot_lmbs
, 0, 0, -1,
393 SPAPR_LMB_FLAGS_RESERVED
|
394 SPAPR_LMB_FLAGS_DRC_INVALID
);
395 QSIMPLEQ_INSERT_TAIL(&drconf_queue
, elem
, entry
);
398 cur_addr
= machine
->device_memory
->base
;
399 for (info
= dimms
; info
; info
= info
->next
) {
400 PCDIMMDeviceInfo
*di
= info
->value
->u
.dimm
.data
;
407 * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
408 * area is marked hotpluggable in the next iteration for the bigger
409 * chunk including the NVDIMM occupied area.
411 if (info
->value
->type
== MEMORY_DEVICE_INFO_KIND_NVDIMM
)
414 /* Entry for hot-pluggable area */
415 if (cur_addr
< addr
) {
416 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
, cur_addr
/ lmb_size
);
418 elem
= spapr_get_drconf_cell((addr
- cur_addr
) / lmb_size
,
419 cur_addr
, spapr_drc_index(drc
), -1, 0);
420 QSIMPLEQ_INSERT_TAIL(&drconf_queue
, elem
, entry
);
425 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
, addr
/ lmb_size
);
427 elem
= spapr_get_drconf_cell(size
/ lmb_size
, addr
,
428 spapr_drc_index(drc
), node
,
429 (SPAPR_LMB_FLAGS_ASSIGNED
|
430 SPAPR_LMB_FLAGS_HOTREMOVABLE
));
431 QSIMPLEQ_INSERT_TAIL(&drconf_queue
, elem
, entry
);
433 cur_addr
= addr
+ size
;
436 /* Entry for remaining hotpluggable area */
437 if (cur_addr
< mem_end
) {
438 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
, cur_addr
/ lmb_size
);
440 elem
= spapr_get_drconf_cell((mem_end
- cur_addr
) / lmb_size
,
441 cur_addr
, spapr_drc_index(drc
), -1, 0);
442 QSIMPLEQ_INSERT_TAIL(&drconf_queue
, elem
, entry
);
446 buf_len
= nr_entries
* sizeof(struct sPAPRDrconfCellV2
) + sizeof(uint32_t);
447 int_buf
= cur_index
= g_malloc0(buf_len
);
448 *(uint32_t *)int_buf
= cpu_to_be32(nr_entries
);
449 cur_index
+= sizeof(nr_entries
);
451 QSIMPLEQ_FOREACH_SAFE(elem
, &drconf_queue
, entry
, next
) {
452 memcpy(cur_index
, &elem
->cell
, sizeof(elem
->cell
));
453 cur_index
+= sizeof(elem
->cell
);
454 QSIMPLEQ_REMOVE(&drconf_queue
, elem
, DrconfCellQueue
, entry
);
458 ret
= fdt_setprop(fdt
, offset
, "ibm,dynamic-memory-v2", int_buf
, buf_len
);
466 static int spapr_dt_dynamic_memory(SpaprMachineState
*spapr
, void *fdt
,
467 int offset
, MemoryDeviceInfoList
*dimms
)
469 MachineState
*machine
= MACHINE(spapr
);
471 uint64_t lmb_size
= SPAPR_MEMORY_BLOCK_SIZE
;
472 uint32_t device_lmb_start
= machine
->device_memory
->base
/ lmb_size
;
473 uint32_t nr_lmbs
= (machine
->device_memory
->base
+
474 memory_region_size(&machine
->device_memory
->mr
)) /
476 uint32_t *int_buf
, *cur_index
, buf_len
;
479 * Allocate enough buffer size to fit in ibm,dynamic-memory
481 buf_len
= (nr_lmbs
* SPAPR_DR_LMB_LIST_ENTRY_SIZE
+ 1) * sizeof(uint32_t);
482 cur_index
= int_buf
= g_malloc0(buf_len
);
483 int_buf
[0] = cpu_to_be32(nr_lmbs
);
485 for (i
= 0; i
< nr_lmbs
; i
++) {
486 uint64_t addr
= i
* lmb_size
;
487 uint32_t *dynamic_memory
= cur_index
;
489 if (i
>= device_lmb_start
) {
492 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
, i
);
495 dynamic_memory
[0] = cpu_to_be32(addr
>> 32);
496 dynamic_memory
[1] = cpu_to_be32(addr
& 0xffffffff);
497 dynamic_memory
[2] = cpu_to_be32(spapr_drc_index(drc
));
498 dynamic_memory
[3] = cpu_to_be32(0); /* reserved */
499 dynamic_memory
[4] = cpu_to_be32(spapr_pc_dimm_node(dimms
, addr
));
500 if (memory_region_present(get_system_memory(), addr
)) {
501 dynamic_memory
[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED
);
503 dynamic_memory
[5] = cpu_to_be32(0);
507 * LMB information for RMA, boot time RAM and gap b/n RAM and
508 * device memory region -- all these are marked as reserved
509 * and as having no valid DRC.
511 dynamic_memory
[0] = cpu_to_be32(addr
>> 32);
512 dynamic_memory
[1] = cpu_to_be32(addr
& 0xffffffff);
513 dynamic_memory
[2] = cpu_to_be32(0);
514 dynamic_memory
[3] = cpu_to_be32(0); /* reserved */
515 dynamic_memory
[4] = cpu_to_be32(-1);
516 dynamic_memory
[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED
|
517 SPAPR_LMB_FLAGS_DRC_INVALID
);
520 cur_index
+= SPAPR_DR_LMB_LIST_ENTRY_SIZE
;
522 ret
= fdt_setprop(fdt
, offset
, "ibm,dynamic-memory", int_buf
, buf_len
);
531 * Adds ibm,dynamic-reconfiguration-memory node.
532 * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
533 * of this device tree node.
535 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState
*spapr
,
538 MachineState
*machine
= MACHINE(spapr
);
540 uint64_t lmb_size
= SPAPR_MEMORY_BLOCK_SIZE
;
541 uint32_t prop_lmb_size
[] = {cpu_to_be32(lmb_size
>> 32),
542 cpu_to_be32(lmb_size
& 0xffffffff)};
543 MemoryDeviceInfoList
*dimms
= NULL
;
546 * Don't create the node if there is no device memory
548 if (machine
->ram_size
== machine
->maxram_size
) {
552 offset
= fdt_add_subnode(fdt
, 0, "ibm,dynamic-reconfiguration-memory");
554 ret
= fdt_setprop(fdt
, offset
, "ibm,lmb-size", prop_lmb_size
,
555 sizeof(prop_lmb_size
));
560 ret
= fdt_setprop_cell(fdt
, offset
, "ibm,memory-flags-mask", 0xff);
565 ret
= fdt_setprop_cell(fdt
, offset
, "ibm,memory-preservation-time", 0x0);
570 /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
571 dimms
= qmp_memory_device_list();
572 if (spapr_ovec_test(spapr
->ov5_cas
, OV5_DRMEM_V2
)) {
573 ret
= spapr_dt_dynamic_memory_v2(spapr
, fdt
, offset
, dimms
);
575 ret
= spapr_dt_dynamic_memory(spapr
, fdt
, offset
, dimms
);
577 qapi_free_MemoryDeviceInfoList(dimms
);
583 ret
= spapr_numa_write_assoc_lookup_arrays(spapr
, fdt
, offset
);
588 static int spapr_dt_memory(SpaprMachineState
*spapr
, void *fdt
)
590 MachineState
*machine
= MACHINE(spapr
);
591 SpaprMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(spapr
);
592 hwaddr mem_start
, node_size
;
593 int i
, nb_nodes
= machine
->numa_state
->num_nodes
;
594 NodeInfo
*nodes
= machine
->numa_state
->nodes
;
596 for (i
= 0, mem_start
= 0; i
< nb_nodes
; ++i
) {
597 if (!nodes
[i
].node_mem
) {
600 if (mem_start
>= machine
->ram_size
) {
603 node_size
= nodes
[i
].node_mem
;
604 if (node_size
> machine
->ram_size
- mem_start
) {
605 node_size
= machine
->ram_size
- mem_start
;
609 /* spapr_machine_init() checks for rma_size <= node0_size
611 spapr_dt_memory_node(spapr
, fdt
, i
, 0, spapr
->rma_size
);
612 mem_start
+= spapr
->rma_size
;
613 node_size
-= spapr
->rma_size
;
615 for ( ; node_size
; ) {
616 hwaddr sizetmp
= pow2floor(node_size
);
618 /* mem_start != 0 here */
619 if (ctzl(mem_start
) < ctzl(sizetmp
)) {
620 sizetmp
= 1ULL << ctzl(mem_start
);
623 spapr_dt_memory_node(spapr
, fdt
, i
, mem_start
, sizetmp
);
624 node_size
-= sizetmp
;
625 mem_start
+= sizetmp
;
629 /* Generate ibm,dynamic-reconfiguration-memory node if required */
630 if (spapr_ovec_test(spapr
->ov5_cas
, OV5_DRCONF_MEMORY
)) {
633 g_assert(smc
->dr_lmb_enabled
);
634 ret
= spapr_dt_dynamic_reconfiguration_memory(spapr
, fdt
);
643 static void spapr_dt_cpu(CPUState
*cs
, void *fdt
, int offset
,
644 SpaprMachineState
*spapr
)
646 MachineState
*ms
= MACHINE(spapr
);
647 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
648 CPUPPCState
*env
= &cpu
->env
;
649 PowerPCCPUClass
*pcc
= POWERPC_CPU_GET_CLASS(cs
);
650 int index
= spapr_get_vcpu_id(cpu
);
651 uint32_t segs
[] = {cpu_to_be32(28), cpu_to_be32(40),
652 0xffffffff, 0xffffffff};
653 uint32_t tbfreq
= kvm_enabled() ? kvmppc_get_tbfreq()
654 : SPAPR_TIMEBASE_FREQ
;
655 uint32_t cpufreq
= kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
656 uint32_t page_sizes_prop
[64];
657 size_t page_sizes_prop_size
;
658 unsigned int smp_threads
= ms
->smp
.threads
;
659 uint32_t vcpus_per_socket
= smp_threads
* ms
->smp
.cores
;
660 uint32_t pft_size_prop
[] = {0, cpu_to_be32(spapr
->htab_shift
)};
661 int compat_smt
= MIN(smp_threads
, ppc_compat_max_vthreads(cpu
));
664 uint32_t radix_AP_encodings
[PPC_PAGE_SIZES_MAX_SZ
];
667 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_CPU
, index
);
669 drc_index
= spapr_drc_index(drc
);
670 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,my-drc-index", drc_index
)));
673 _FDT((fdt_setprop_cell(fdt
, offset
, "reg", index
)));
674 _FDT((fdt_setprop_string(fdt
, offset
, "device_type", "cpu")));
676 _FDT((fdt_setprop_cell(fdt
, offset
, "cpu-version", env
->spr
[SPR_PVR
])));
677 _FDT((fdt_setprop_cell(fdt
, offset
, "d-cache-block-size",
678 env
->dcache_line_size
)));
679 _FDT((fdt_setprop_cell(fdt
, offset
, "d-cache-line-size",
680 env
->dcache_line_size
)));
681 _FDT((fdt_setprop_cell(fdt
, offset
, "i-cache-block-size",
682 env
->icache_line_size
)));
683 _FDT((fdt_setprop_cell(fdt
, offset
, "i-cache-line-size",
684 env
->icache_line_size
)));
686 if (pcc
->l1_dcache_size
) {
687 _FDT((fdt_setprop_cell(fdt
, offset
, "d-cache-size",
688 pcc
->l1_dcache_size
)));
690 warn_report("Unknown L1 dcache size for cpu");
692 if (pcc
->l1_icache_size
) {
693 _FDT((fdt_setprop_cell(fdt
, offset
, "i-cache-size",
694 pcc
->l1_icache_size
)));
696 warn_report("Unknown L1 icache size for cpu");
699 _FDT((fdt_setprop_cell(fdt
, offset
, "timebase-frequency", tbfreq
)));
700 _FDT((fdt_setprop_cell(fdt
, offset
, "clock-frequency", cpufreq
)));
701 _FDT((fdt_setprop_cell(fdt
, offset
, "slb-size", cpu
->hash64_opts
->slb_size
)));
702 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,slb-size", cpu
->hash64_opts
->slb_size
)));
703 _FDT((fdt_setprop_string(fdt
, offset
, "status", "okay")));
704 _FDT((fdt_setprop(fdt
, offset
, "64-bit", NULL
, 0)));
706 if (ppc_has_spr(cpu
, SPR_PURR
)) {
707 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,purr", 1)));
709 if (ppc_has_spr(cpu
, SPR_PURR
)) {
710 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,spurr", 1)));
713 if (ppc_hash64_has(cpu
, PPC_HASH64_1TSEG
)) {
714 _FDT((fdt_setprop(fdt
, offset
, "ibm,processor-segment-sizes",
715 segs
, sizeof(segs
))));
718 /* Advertise VSX (vector extensions) if available
719 * 1 == VMX / Altivec available
722 * Only CPUs for which we create core types in spapr_cpu_core.c
723 * are possible, and all of those have VMX */
724 if (spapr_get_cap(spapr
, SPAPR_CAP_VSX
) != 0) {
725 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,vmx", 2)));
727 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,vmx", 1)));
730 /* Advertise DFP (Decimal Floating Point) if available
731 * 0 / no property == no DFP
732 * 1 == DFP available */
733 if (spapr_get_cap(spapr
, SPAPR_CAP_DFP
) != 0) {
734 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,dfp", 1)));
737 page_sizes_prop_size
= ppc_create_page_sizes_prop(cpu
, page_sizes_prop
,
738 sizeof(page_sizes_prop
));
739 if (page_sizes_prop_size
) {
740 _FDT((fdt_setprop(fdt
, offset
, "ibm,segment-page-sizes",
741 page_sizes_prop
, page_sizes_prop_size
)));
744 spapr_dt_pa_features(spapr
, cpu
, fdt
, offset
);
746 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,chip-id",
747 cs
->cpu_index
/ vcpus_per_socket
)));
749 _FDT((fdt_setprop(fdt
, offset
, "ibm,pft-size",
750 pft_size_prop
, sizeof(pft_size_prop
))));
752 if (ms
->numa_state
->num_nodes
> 1) {
753 _FDT(spapr_numa_fixup_cpu_dt(spapr
, fdt
, offset
, cpu
));
756 _FDT(spapr_fixup_cpu_smt_dt(fdt
, offset
, cpu
, compat_smt
));
758 if (pcc
->radix_page_info
) {
759 for (i
= 0; i
< pcc
->radix_page_info
->count
; i
++) {
760 radix_AP_encodings
[i
] =
761 cpu_to_be32(pcc
->radix_page_info
->entries
[i
]);
763 _FDT((fdt_setprop(fdt
, offset
, "ibm,processor-radix-AP-encodings",
765 pcc
->radix_page_info
->count
*
766 sizeof(radix_AP_encodings
[0]))));
770 * We set this property to let the guest know that it can use the large
771 * decrementer and its width in bits.
773 if (spapr_get_cap(spapr
, SPAPR_CAP_LARGE_DECREMENTER
) != SPAPR_CAP_OFF
)
774 _FDT((fdt_setprop_u32(fdt
, offset
, "ibm,dec-bits",
775 pcc
->lrg_decr_bits
)));
778 static void spapr_dt_cpus(void *fdt
, SpaprMachineState
*spapr
)
786 cpus_offset
= fdt_add_subnode(fdt
, 0, "cpus");
788 _FDT((fdt_setprop_cell(fdt
, cpus_offset
, "#address-cells", 0x1)));
789 _FDT((fdt_setprop_cell(fdt
, cpus_offset
, "#size-cells", 0x0)));
792 * We walk the CPUs in reverse order to ensure that CPU DT nodes
793 * created by fdt_add_subnode() end up in the right order in FDT
794 * for the guest kernel the enumerate the CPUs correctly.
796 * The CPU list cannot be traversed in reverse order, so we need
802 rev
= g_renew(CPUState
*, rev
, n_cpus
+ 1);
806 for (i
= n_cpus
- 1; i
>= 0; i
--) {
807 CPUState
*cs
= rev
[i
];
808 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
809 int index
= spapr_get_vcpu_id(cpu
);
810 DeviceClass
*dc
= DEVICE_GET_CLASS(cs
);
811 g_autofree
char *nodename
= NULL
;
814 if (!spapr_is_thread0_in_vcore(spapr
, cpu
)) {
818 nodename
= g_strdup_printf("%s@%x", dc
->fw_name
, index
);
819 offset
= fdt_add_subnode(fdt
, cpus_offset
, nodename
);
821 spapr_dt_cpu(cs
, fdt
, offset
, spapr
);
827 static int spapr_dt_rng(void *fdt
)
832 node
= qemu_fdt_add_subnode(fdt
, "/ibm,platform-facilities");
836 ret
= fdt_setprop_string(fdt
, node
, "device_type",
837 "ibm,platform-facilities");
838 ret
|= fdt_setprop_cell(fdt
, node
, "#address-cells", 0x1);
839 ret
|= fdt_setprop_cell(fdt
, node
, "#size-cells", 0x0);
841 node
= fdt_add_subnode(fdt
, node
, "ibm,random-v1");
845 ret
|= fdt_setprop_string(fdt
, node
, "compatible", "ibm,random");
850 static void spapr_dt_rtas(SpaprMachineState
*spapr
, void *fdt
)
852 MachineState
*ms
= MACHINE(spapr
);
854 GString
*hypertas
= g_string_sized_new(256);
855 GString
*qemu_hypertas
= g_string_sized_new(256);
856 uint64_t max_device_addr
= MACHINE(spapr
)->device_memory
->base
+
857 memory_region_size(&MACHINE(spapr
)->device_memory
->mr
);
858 uint32_t lrdr_capacity
[] = {
859 cpu_to_be32(max_device_addr
>> 32),
860 cpu_to_be32(max_device_addr
& 0xffffffff),
861 cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE
>> 32),
862 cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE
& 0xffffffff),
863 cpu_to_be32(ms
->smp
.max_cpus
/ ms
->smp
.threads
),
866 _FDT(rtas
= fdt_add_subnode(fdt
, 0, "rtas"));
869 add_str(hypertas
, "hcall-pft");
870 add_str(hypertas
, "hcall-term");
871 add_str(hypertas
, "hcall-dabr");
872 add_str(hypertas
, "hcall-interrupt");
873 add_str(hypertas
, "hcall-tce");
874 add_str(hypertas
, "hcall-vio");
875 add_str(hypertas
, "hcall-splpar");
876 add_str(hypertas
, "hcall-join");
877 add_str(hypertas
, "hcall-bulk");
878 add_str(hypertas
, "hcall-set-mode");
879 add_str(hypertas
, "hcall-sprg0");
880 add_str(hypertas
, "hcall-copy");
881 add_str(hypertas
, "hcall-debug");
882 add_str(hypertas
, "hcall-vphn");
883 add_str(qemu_hypertas
, "hcall-memop1");
885 if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
886 add_str(hypertas
, "hcall-multi-tce");
889 if (spapr
->resize_hpt
!= SPAPR_RESIZE_HPT_DISABLED
) {
890 add_str(hypertas
, "hcall-hpt-resize");
893 _FDT(fdt_setprop(fdt
, rtas
, "ibm,hypertas-functions",
894 hypertas
->str
, hypertas
->len
));
895 g_string_free(hypertas
, TRUE
);
896 _FDT(fdt_setprop(fdt
, rtas
, "qemu,hypertas-functions",
897 qemu_hypertas
->str
, qemu_hypertas
->len
));
898 g_string_free(qemu_hypertas
, TRUE
);
900 spapr_numa_write_rtas_dt(spapr
, fdt
, rtas
);
903 * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
904 * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
906 * The system reset requirements are driven by existing Linux and PowerVM
907 * implementation which (contrary to PAPR) saves r3 in the error log
908 * structure like machine check, so Linux expects to find the saved r3
909 * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
910 * does not look at the error value).
912 * System reset interrupts are not subject to interlock like machine
913 * check, so this memory area could be corrupted if the sreset is
914 * interrupted by a machine check (or vice versa) if it was shared. To
915 * prevent this, system reset uses per-CPU areas for the sreset save
916 * area. A system reset that interrupts a system reset handler could
917 * still overwrite this area, but Linux doesn't try to recover in that
920 * The extra 8 bytes is required because Linux's FWNMI error log check
923 _FDT(fdt_setprop_cell(fdt
, rtas
, "rtas-size", RTAS_ERROR_LOG_MAX
+
924 ms
->smp
.max_cpus
* sizeof(uint64_t)*2 + sizeof(uint64_t)));
925 _FDT(fdt_setprop_cell(fdt
, rtas
, "rtas-error-log-max",
926 RTAS_ERROR_LOG_MAX
));
927 _FDT(fdt_setprop_cell(fdt
, rtas
, "rtas-event-scan-rate",
928 RTAS_EVENT_SCAN_RATE
));
930 g_assert(msi_nonbroken
);
931 _FDT(fdt_setprop(fdt
, rtas
, "ibm,change-msix-capable", NULL
, 0));
934 * According to PAPR, rtas ibm,os-term does not guarantee a return
935 * back to the guest cpu.
937 * While an additional ibm,extended-os-term property indicates
938 * that rtas call return will always occur. Set this property.
940 _FDT(fdt_setprop(fdt
, rtas
, "ibm,extended-os-term", NULL
, 0));
942 _FDT(fdt_setprop(fdt
, rtas
, "ibm,lrdr-capacity",
943 lrdr_capacity
, sizeof(lrdr_capacity
)));
945 spapr_dt_rtas_tokens(fdt
, rtas
);
949 * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
950 * and the XIVE features that the guest may request and thus the valid
951 * values for bytes 23..26 of option vector 5:
953 static void spapr_dt_ov5_platform_support(SpaprMachineState
*spapr
, void *fdt
,
956 PowerPCCPU
*first_ppc_cpu
= POWERPC_CPU(first_cpu
);
959 23, 0x00, /* XICS / XIVE mode */
960 24, 0x00, /* Hash/Radix, filled in below. */
961 25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
962 26, 0x40, /* Radix options: GTSE == yes. */
965 if (spapr
->irq
->xics
&& spapr
->irq
->xive
) {
966 val
[1] = SPAPR_OV5_XIVE_BOTH
;
967 } else if (spapr
->irq
->xive
) {
968 val
[1] = SPAPR_OV5_XIVE_EXPLOIT
;
970 assert(spapr
->irq
->xics
);
971 val
[1] = SPAPR_OV5_XIVE_LEGACY
;
974 if (!ppc_check_compat(first_ppc_cpu
, CPU_POWERPC_LOGICAL_3_00
, 0,
975 first_ppc_cpu
->compat_pvr
)) {
977 * If we're in a pre POWER9 compat mode then the guest should
978 * do hash and use the legacy interrupt mode
980 val
[1] = SPAPR_OV5_XIVE_LEGACY
; /* XICS */
981 val
[3] = 0x00; /* Hash */
982 spapr_check_mmu_mode(false);
983 } else if (kvm_enabled()) {
984 if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
985 val
[3] = 0x80; /* OV5_MMU_BOTH */
986 } else if (kvmppc_has_cap_mmu_radix()) {
987 val
[3] = 0x40; /* OV5_MMU_RADIX_300 */
989 val
[3] = 0x00; /* Hash */
992 /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
995 _FDT(fdt_setprop(fdt
, chosen
, "ibm,arch-vec-5-platform-support",
999 static void spapr_dt_chosen(SpaprMachineState
*spapr
, void *fdt
, bool reset
)
1001 MachineState
*machine
= MACHINE(spapr
);
1002 SpaprMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(machine
);
1005 _FDT(chosen
= fdt_add_subnode(fdt
, 0, "chosen"));
1008 const char *boot_device
= machine
->boot_order
;
1009 char *stdout_path
= spapr_vio_stdout_path(spapr
->vio_bus
);
1011 char *bootlist
= get_boot_devices_list(&cb
);
1013 if (machine
->kernel_cmdline
&& machine
->kernel_cmdline
[0]) {
1014 _FDT(fdt_setprop_string(fdt
, chosen
, "bootargs",
1015 machine
->kernel_cmdline
));
1018 if (spapr
->initrd_size
) {
1019 _FDT(fdt_setprop_cell(fdt
, chosen
, "linux,initrd-start",
1020 spapr
->initrd_base
));
1021 _FDT(fdt_setprop_cell(fdt
, chosen
, "linux,initrd-end",
1022 spapr
->initrd_base
+ spapr
->initrd_size
));
1025 if (spapr
->kernel_size
) {
1026 uint64_t kprop
[2] = { cpu_to_be64(spapr
->kernel_addr
),
1027 cpu_to_be64(spapr
->kernel_size
) };
1029 _FDT(fdt_setprop(fdt
, chosen
, "qemu,boot-kernel",
1030 &kprop
, sizeof(kprop
)));
1031 if (spapr
->kernel_le
) {
1032 _FDT(fdt_setprop(fdt
, chosen
, "qemu,boot-kernel-le", NULL
, 0));
1036 _FDT((fdt_setprop_cell(fdt
, chosen
, "qemu,boot-menu", boot_menu
)));
1038 _FDT(fdt_setprop_cell(fdt
, chosen
, "qemu,graphic-width", graphic_width
));
1039 _FDT(fdt_setprop_cell(fdt
, chosen
, "qemu,graphic-height", graphic_height
));
1040 _FDT(fdt_setprop_cell(fdt
, chosen
, "qemu,graphic-depth", graphic_depth
));
1042 if (cb
&& bootlist
) {
1045 for (i
= 0; i
< cb
; i
++) {
1046 if (bootlist
[i
] == '\n') {
1050 _FDT(fdt_setprop_string(fdt
, chosen
, "qemu,boot-list", bootlist
));
1053 if (boot_device
&& strlen(boot_device
)) {
1054 _FDT(fdt_setprop_string(fdt
, chosen
, "qemu,boot-device", boot_device
));
1057 if (!spapr
->has_graphics
&& stdout_path
) {
1059 * "linux,stdout-path" and "stdout" properties are
1060 * deprecated by linux kernel. New platforms should only
1061 * use the "stdout-path" property. Set the new property
1062 * and continue using older property to remain compatible
1063 * with the existing firmware.
1065 _FDT(fdt_setprop_string(fdt
, chosen
, "linux,stdout-path", stdout_path
));
1066 _FDT(fdt_setprop_string(fdt
, chosen
, "stdout-path", stdout_path
));
1070 * We can deal with BAR reallocation just fine, advertise it
1073 if (smc
->linux_pci_probe
) {
1074 _FDT(fdt_setprop_cell(fdt
, chosen
, "linux,pci-probe-only", 0));
1077 spapr_dt_ov5_platform_support(spapr
, fdt
, chosen
);
1079 g_free(stdout_path
);
1083 _FDT(spapr_dt_ovec(fdt
, chosen
, spapr
->ov5_cas
, "ibm,architecture-vec-5"));
1086 static void spapr_dt_hypervisor(SpaprMachineState
*spapr
, void *fdt
)
1088 /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1089 * KVM to work under pHyp with some guest co-operation */
1091 uint8_t hypercall
[16];
1093 _FDT(hypervisor
= fdt_add_subnode(fdt
, 0, "hypervisor"));
1094 /* indicate KVM hypercall interface */
1095 _FDT(fdt_setprop_string(fdt
, hypervisor
, "compatible", "linux,kvm"));
1096 if (kvmppc_has_cap_fixup_hcalls()) {
1098 * Older KVM versions with older guest kernels were broken
1099 * with the magic page, don't allow the guest to map it.
1101 if (!kvmppc_get_hypercall(first_cpu
->env_ptr
, hypercall
,
1102 sizeof(hypercall
))) {
1103 _FDT(fdt_setprop(fdt
, hypervisor
, "hcall-instructions",
1104 hypercall
, sizeof(hypercall
)));
1109 void *spapr_build_fdt(SpaprMachineState
*spapr
, bool reset
, size_t space
)
1111 MachineState
*machine
= MACHINE(spapr
);
1112 MachineClass
*mc
= MACHINE_GET_CLASS(machine
);
1113 SpaprMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(machine
);
1114 uint32_t root_drc_type_mask
= 0;
1120 fdt
= g_malloc0(space
);
1121 _FDT((fdt_create_empty_tree(fdt
, space
)));
1124 _FDT(fdt_setprop_string(fdt
, 0, "device_type", "chrp"));
1125 _FDT(fdt_setprop_string(fdt
, 0, "model", "IBM pSeries (emulated by qemu)"));
1126 _FDT(fdt_setprop_string(fdt
, 0, "compatible", "qemu,pseries"));
1128 /* Guest UUID & Name*/
1129 buf
= qemu_uuid_unparse_strdup(&qemu_uuid
);
1130 _FDT(fdt_setprop_string(fdt
, 0, "vm,uuid", buf
));
1131 if (qemu_uuid_set
) {
1132 _FDT(fdt_setprop_string(fdt
, 0, "system-id", buf
));
1136 if (qemu_get_vm_name()) {
1137 _FDT(fdt_setprop_string(fdt
, 0, "ibm,partition-name",
1138 qemu_get_vm_name()));
1141 /* Host Model & Serial Number */
1142 if (spapr
->host_model
) {
1143 _FDT(fdt_setprop_string(fdt
, 0, "host-model", spapr
->host_model
));
1144 } else if (smc
->broken_host_serial_model
&& kvmppc_get_host_model(&buf
)) {
1145 _FDT(fdt_setprop_string(fdt
, 0, "host-model", buf
));
1149 if (spapr
->host_serial
) {
1150 _FDT(fdt_setprop_string(fdt
, 0, "host-serial", spapr
->host_serial
));
1151 } else if (smc
->broken_host_serial_model
&& kvmppc_get_host_serial(&buf
)) {
1152 _FDT(fdt_setprop_string(fdt
, 0, "host-serial", buf
));
1156 _FDT(fdt_setprop_cell(fdt
, 0, "#address-cells", 2));
1157 _FDT(fdt_setprop_cell(fdt
, 0, "#size-cells", 2));
1159 /* /interrupt controller */
1160 spapr_irq_dt(spapr
, spapr_max_server_number(spapr
), fdt
, PHANDLE_INTC
);
1162 ret
= spapr_dt_memory(spapr
, fdt
);
1164 error_report("couldn't setup memory nodes in fdt");
1169 spapr_dt_vdevice(spapr
->vio_bus
, fdt
);
1171 if (object_resolve_path_type("", TYPE_SPAPR_RNG
, NULL
)) {
1172 ret
= spapr_dt_rng(fdt
);
1174 error_report("could not set up rng device in the fdt");
1179 QLIST_FOREACH(phb
, &spapr
->phbs
, list
) {
1180 ret
= spapr_dt_phb(spapr
, phb
, PHANDLE_INTC
, fdt
, NULL
);
1182 error_report("couldn't setup PCI devices in fdt");
1187 spapr_dt_cpus(fdt
, spapr
);
1189 /* ibm,drc-indexes and friends */
1190 if (smc
->dr_lmb_enabled
) {
1191 root_drc_type_mask
|= SPAPR_DR_CONNECTOR_TYPE_LMB
;
1193 if (smc
->dr_phb_enabled
) {
1194 root_drc_type_mask
|= SPAPR_DR_CONNECTOR_TYPE_PHB
;
1196 if (mc
->nvdimm_supported
) {
1197 root_drc_type_mask
|= SPAPR_DR_CONNECTOR_TYPE_PMEM
;
1199 if (root_drc_type_mask
) {
1200 _FDT(spapr_dt_drc(fdt
, 0, NULL
, root_drc_type_mask
));
1203 if (mc
->has_hotpluggable_cpus
) {
1204 int offset
= fdt_path_offset(fdt
, "/cpus");
1205 ret
= spapr_dt_drc(fdt
, offset
, NULL
, SPAPR_DR_CONNECTOR_TYPE_CPU
);
1207 error_report("Couldn't set up CPU DR device tree properties");
1212 /* /event-sources */
1213 spapr_dt_events(spapr
, fdt
);
1216 spapr_dt_rtas(spapr
, fdt
);
1219 spapr_dt_chosen(spapr
, fdt
, reset
);
1222 if (kvm_enabled()) {
1223 spapr_dt_hypervisor(spapr
, fdt
);
1226 /* Build memory reserve map */
1228 if (spapr
->kernel_size
) {
1229 _FDT((fdt_add_mem_rsv(fdt
, spapr
->kernel_addr
,
1230 spapr
->kernel_size
)));
1232 if (spapr
->initrd_size
) {
1233 _FDT((fdt_add_mem_rsv(fdt
, spapr
->initrd_base
,
1234 spapr
->initrd_size
)));
1238 /* NVDIMM devices */
1239 if (mc
->nvdimm_supported
) {
1240 spapr_dt_persistent_memory(spapr
, fdt
);
1246 static uint64_t translate_kernel_address(void *opaque
, uint64_t addr
)
1248 SpaprMachineState
*spapr
= opaque
;
1250 return (addr
& 0x0fffffff) + spapr
->kernel_addr
;
1253 static void emulate_spapr_hypercall(PPCVirtualHypervisor
*vhyp
,
1256 CPUPPCState
*env
= &cpu
->env
;
1258 /* The TCG path should also be holding the BQL at this point */
1259 g_assert(qemu_mutex_iothread_locked());
1262 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1263 env
->gpr
[3] = H_PRIVILEGE
;
1265 env
->gpr
[3] = spapr_hypercall(cpu
, env
->gpr
[3], &env
->gpr
[4]);
1269 struct LPCRSyncState
{
1274 static void do_lpcr_sync(CPUState
*cs
, run_on_cpu_data arg
)
1276 struct LPCRSyncState
*s
= arg
.host_ptr
;
1277 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
1278 CPUPPCState
*env
= &cpu
->env
;
1281 cpu_synchronize_state(cs
);
1282 lpcr
= env
->spr
[SPR_LPCR
];
1285 ppc_store_lpcr(cpu
, lpcr
);
1288 void spapr_set_all_lpcrs(target_ulong value
, target_ulong mask
)
1291 struct LPCRSyncState s
= {
1296 run_on_cpu(cs
, do_lpcr_sync
, RUN_ON_CPU_HOST_PTR(&s
));
1300 static void spapr_get_pate(PPCVirtualHypervisor
*vhyp
, ppc_v3_pate_t
*entry
)
1302 SpaprMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1304 /* Copy PATE1:GR into PATE0:HR */
1305 entry
->dw0
= spapr
->patb_entry
& PATE0_HR
;
1306 entry
->dw1
= spapr
->patb_entry
;
1309 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1310 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1311 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1312 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1313 #define DIRTY_HPTE(_hpte) ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1316 * Get the fd to access the kernel htab, re-opening it if necessary
1318 static int get_htab_fd(SpaprMachineState
*spapr
)
1320 Error
*local_err
= NULL
;
1322 if (spapr
->htab_fd
>= 0) {
1323 return spapr
->htab_fd
;
1326 spapr
->htab_fd
= kvmppc_get_htab_fd(false, 0, &local_err
);
1327 if (spapr
->htab_fd
< 0) {
1328 error_report_err(local_err
);
1331 return spapr
->htab_fd
;
1334 void close_htab_fd(SpaprMachineState
*spapr
)
1336 if (spapr
->htab_fd
>= 0) {
1337 close(spapr
->htab_fd
);
1339 spapr
->htab_fd
= -1;
1342 static hwaddr
spapr_hpt_mask(PPCVirtualHypervisor
*vhyp
)
1344 SpaprMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1346 return HTAB_SIZE(spapr
) / HASH_PTEG_SIZE_64
- 1;
1349 static target_ulong
spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor
*vhyp
)
1351 SpaprMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1353 assert(kvm_enabled());
1359 return (target_ulong
)(uintptr_t)spapr
->htab
| (spapr
->htab_shift
- 18);
1362 static const ppc_hash_pte64_t
*spapr_map_hptes(PPCVirtualHypervisor
*vhyp
,
1365 SpaprMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1366 hwaddr pte_offset
= ptex
* HASH_PTE_SIZE_64
;
1370 * HTAB is controlled by KVM. Fetch into temporary buffer
1372 ppc_hash_pte64_t
*hptes
= g_malloc(n
* HASH_PTE_SIZE_64
);
1373 kvmppc_read_hptes(hptes
, ptex
, n
);
1378 * HTAB is controlled by QEMU. Just point to the internally
1381 return (const ppc_hash_pte64_t
*)(spapr
->htab
+ pte_offset
);
1384 static void spapr_unmap_hptes(PPCVirtualHypervisor
*vhyp
,
1385 const ppc_hash_pte64_t
*hptes
,
1388 SpaprMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1391 g_free((void *)hptes
);
1394 /* Nothing to do for qemu managed HPT */
1397 void spapr_store_hpte(PowerPCCPU
*cpu
, hwaddr ptex
,
1398 uint64_t pte0
, uint64_t pte1
)
1400 SpaprMachineState
*spapr
= SPAPR_MACHINE(cpu
->vhyp
);
1401 hwaddr offset
= ptex
* HASH_PTE_SIZE_64
;
1404 kvmppc_write_hpte(ptex
, pte0
, pte1
);
1406 if (pte0
& HPTE64_V_VALID
) {
1407 stq_p(spapr
->htab
+ offset
+ HASH_PTE_SIZE_64
/ 2, pte1
);
1409 * When setting valid, we write PTE1 first. This ensures
1410 * proper synchronization with the reading code in
1411 * ppc_hash64_pteg_search()
1414 stq_p(spapr
->htab
+ offset
, pte0
);
1416 stq_p(spapr
->htab
+ offset
, pte0
);
1418 * When clearing it we set PTE0 first. This ensures proper
1419 * synchronization with the reading code in
1420 * ppc_hash64_pteg_search()
1423 stq_p(spapr
->htab
+ offset
+ HASH_PTE_SIZE_64
/ 2, pte1
);
1428 static void spapr_hpte_set_c(PPCVirtualHypervisor
*vhyp
, hwaddr ptex
,
1431 hwaddr offset
= ptex
* HASH_PTE_SIZE_64
+ 15;
1432 SpaprMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1435 /* There should always be a hash table when this is called */
1436 error_report("spapr_hpte_set_c called with no hash table !");
1440 /* The HW performs a non-atomic byte update */
1441 stb_p(spapr
->htab
+ offset
, (pte1
& 0xff) | 0x80);
1444 static void spapr_hpte_set_r(PPCVirtualHypervisor
*vhyp
, hwaddr ptex
,
1447 hwaddr offset
= ptex
* HASH_PTE_SIZE_64
+ 14;
1448 SpaprMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1451 /* There should always be a hash table when this is called */
1452 error_report("spapr_hpte_set_r called with no hash table !");
1456 /* The HW performs a non-atomic byte update */
1457 stb_p(spapr
->htab
+ offset
, ((pte1
>> 8) & 0xff) | 0x01);
1460 int spapr_hpt_shift_for_ramsize(uint64_t ramsize
)
1464 /* We aim for a hash table of size 1/128 the size of RAM (rounded
1465 * up). The PAPR recommendation is actually 1/64 of RAM size, but
1466 * that's much more than is needed for Linux guests */
1467 shift
= ctz64(pow2ceil(ramsize
)) - 7;
1468 shift
= MAX(shift
, 18); /* Minimum architected size */
1469 shift
= MIN(shift
, 46); /* Maximum architected size */
1473 void spapr_free_hpt(SpaprMachineState
*spapr
)
1475 g_free(spapr
->htab
);
1477 spapr
->htab_shift
= 0;
1478 close_htab_fd(spapr
);
1481 int spapr_reallocate_hpt(SpaprMachineState
*spapr
, int shift
, Error
**errp
)
1486 /* Clean up any HPT info from a previous boot */
1487 spapr_free_hpt(spapr
);
1489 rc
= kvmppc_reset_htab(shift
);
1491 if (rc
== -EOPNOTSUPP
) {
1492 error_setg(errp
, "HPT not supported in nested guests");
1497 /* kernel-side HPT needed, but couldn't allocate one */
1498 error_setg_errno(errp
, errno
, "Failed to allocate KVM HPT of order %d",
1500 error_append_hint(errp
, "Try smaller maxmem?\n");
1502 } else if (rc
> 0) {
1503 /* kernel-side HPT allocated */
1506 "Requested order %d HPT, but kernel allocated order %ld",
1508 error_append_hint(errp
, "Try smaller maxmem?\n");
1512 spapr
->htab_shift
= shift
;
1515 /* kernel-side HPT not needed, allocate in userspace instead */
1516 size_t size
= 1ULL << shift
;
1519 spapr
->htab
= qemu_memalign(size
, size
);
1520 memset(spapr
->htab
, 0, size
);
1521 spapr
->htab_shift
= shift
;
1523 for (i
= 0; i
< size
/ HASH_PTE_SIZE_64
; i
++) {
1524 DIRTY_HPTE(HPTE(spapr
->htab
, i
));
1527 /* We're setting up a hash table, so that means we're not radix */
1528 spapr
->patb_entry
= 0;
1529 spapr_set_all_lpcrs(0, LPCR_HR
| LPCR_UPRT
);
1533 void spapr_setup_hpt(SpaprMachineState
*spapr
)
1537 if (spapr
->resize_hpt
== SPAPR_RESIZE_HPT_DISABLED
) {
1538 hpt_shift
= spapr_hpt_shift_for_ramsize(MACHINE(spapr
)->maxram_size
);
1540 uint64_t current_ram_size
;
1542 current_ram_size
= MACHINE(spapr
)->ram_size
+ get_plugged_memory_size();
1543 hpt_shift
= spapr_hpt_shift_for_ramsize(current_ram_size
);
1545 spapr_reallocate_hpt(spapr
, hpt_shift
, &error_fatal
);
1547 if (kvm_enabled()) {
1548 hwaddr vrma_limit
= kvmppc_vrma_limit(spapr
->htab_shift
);
1550 /* Check our RMA fits in the possible VRMA */
1551 if (vrma_limit
< spapr
->rma_size
) {
1552 error_report("Unable to create %" HWADDR_PRIu
1553 "MiB RMA (VRMA only allows %" HWADDR_PRIu
"MiB",
1554 spapr
->rma_size
/ MiB
, vrma_limit
/ MiB
);
1560 void spapr_check_mmu_mode(bool guest_radix
)
1563 if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1564 error_report("Guest requested unavailable MMU mode (radix).");
1568 if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1569 && !kvmppc_has_cap_mmu_hash_v3()) {
1570 error_report("Guest requested unavailable MMU mode (hash).");
1576 static void spapr_machine_reset(MachineState
*machine
)
1578 SpaprMachineState
*spapr
= SPAPR_MACHINE(machine
);
1579 PowerPCCPU
*first_ppc_cpu
;
1584 pef_kvm_reset(machine
->cgs
, &error_fatal
);
1585 spapr_caps_apply(spapr
);
1587 first_ppc_cpu
= POWERPC_CPU(first_cpu
);
1588 if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1589 ppc_type_check_compat(machine
->cpu_type
, CPU_POWERPC_LOGICAL_3_00
, 0,
1590 spapr
->max_compat_pvr
)) {
1592 * If using KVM with radix mode available, VCPUs can be started
1593 * without a HPT because KVM will start them in radix mode.
1594 * Set the GR bit in PATE so that we know there is no HPT.
1596 spapr
->patb_entry
= PATE1_GR
;
1597 spapr_set_all_lpcrs(LPCR_HR
| LPCR_UPRT
, LPCR_HR
| LPCR_UPRT
);
1599 spapr_setup_hpt(spapr
);
1602 qemu_devices_reset();
1604 spapr_ovec_cleanup(spapr
->ov5_cas
);
1605 spapr
->ov5_cas
= spapr_ovec_new();
1607 ppc_set_compat_all(spapr
->max_compat_pvr
, &error_fatal
);
1610 * This is fixing some of the default configuration of the XIVE
1611 * devices. To be called after the reset of the machine devices.
1613 spapr_irq_reset(spapr
, &error_fatal
);
1616 * There is no CAS under qtest. Simulate one to please the code that
1617 * depends on spapr->ov5_cas. This is especially needed to test device
1618 * unplug, so we do that before resetting the DRCs.
1620 if (qtest_enabled()) {
1621 spapr_ovec_cleanup(spapr
->ov5_cas
);
1622 spapr
->ov5_cas
= spapr_ovec_clone(spapr
->ov5
);
1625 /* DRC reset may cause a device to be unplugged. This will cause troubles
1626 * if this device is used by another device (eg, a running vhost backend
1627 * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1628 * situations, we reset DRCs after all devices have been reset.
1630 spapr_drc_reset_all(spapr
);
1632 spapr_clear_pending_events(spapr
);
1635 * We place the device tree just below either the top of the RMA,
1636 * or just below 2GB, whichever is lower, so that it can be
1637 * processed with 32-bit real mode code if necessary
1639 fdt_addr
= MIN(spapr
->rma_size
, FDT_MAX_ADDR
) - FDT_MAX_SIZE
;
1641 fdt
= spapr_build_fdt(spapr
, true, FDT_MAX_SIZE
);
1645 /* Should only fail if we've built a corrupted tree */
1649 qemu_fdt_dumpdtb(fdt
, fdt_totalsize(fdt
));
1650 cpu_physical_memory_write(fdt_addr
, fdt
, fdt_totalsize(fdt
));
1651 g_free(spapr
->fdt_blob
);
1652 spapr
->fdt_size
= fdt_totalsize(fdt
);
1653 spapr
->fdt_initial_size
= spapr
->fdt_size
;
1654 spapr
->fdt_blob
= fdt
;
1656 /* Set up the entry state */
1657 spapr_cpu_set_entry_state(first_ppc_cpu
, SPAPR_ENTRY_POINT
, 0, fdt_addr
, 0);
1658 first_ppc_cpu
->env
.gpr
[5] = 0;
1660 spapr
->fwnmi_system_reset_addr
= -1;
1661 spapr
->fwnmi_machine_check_addr
= -1;
1662 spapr
->fwnmi_machine_check_interlock
= -1;
1664 /* Signal all vCPUs waiting on this condition */
1665 qemu_cond_broadcast(&spapr
->fwnmi_machine_check_interlock_cond
);
1667 migrate_del_blocker(spapr
->fwnmi_migration_blocker
);
1670 static void spapr_create_nvram(SpaprMachineState
*spapr
)
1672 DeviceState
*dev
= qdev_new("spapr-nvram");
1673 DriveInfo
*dinfo
= drive_get(IF_PFLASH
, 0, 0);
1676 qdev_prop_set_drive_err(dev
, "drive", blk_by_legacy_dinfo(dinfo
),
1680 qdev_realize_and_unref(dev
, &spapr
->vio_bus
->bus
, &error_fatal
);
1682 spapr
->nvram
= (struct SpaprNvram
*)dev
;
1685 static void spapr_rtc_create(SpaprMachineState
*spapr
)
1687 object_initialize_child_with_props(OBJECT(spapr
), "rtc", &spapr
->rtc
,
1688 sizeof(spapr
->rtc
), TYPE_SPAPR_RTC
,
1689 &error_fatal
, NULL
);
1690 qdev_realize(DEVICE(&spapr
->rtc
), NULL
, &error_fatal
);
1691 object_property_add_alias(OBJECT(spapr
), "rtc-time", OBJECT(&spapr
->rtc
),
1695 /* Returns whether we want to use VGA or not */
1696 static bool spapr_vga_init(PCIBus
*pci_bus
, Error
**errp
)
1698 switch (vga_interface_type
) {
1706 return pci_vga_init(pci_bus
) != NULL
;
1709 "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1714 static int spapr_pre_load(void *opaque
)
1718 rc
= spapr_caps_pre_load(opaque
);
1726 static int spapr_post_load(void *opaque
, int version_id
)
1728 SpaprMachineState
*spapr
= (SpaprMachineState
*)opaque
;
1731 err
= spapr_caps_post_migration(spapr
);
1737 * In earlier versions, there was no separate qdev for the PAPR
1738 * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1739 * So when migrating from those versions, poke the incoming offset
1740 * value into the RTC device
1742 if (version_id
< 3) {
1743 err
= spapr_rtc_import_offset(&spapr
->rtc
, spapr
->rtc_offset
);
1749 if (kvm_enabled() && spapr
->patb_entry
) {
1750 PowerPCCPU
*cpu
= POWERPC_CPU(first_cpu
);
1751 bool radix
= !!(spapr
->patb_entry
& PATE1_GR
);
1752 bool gtse
= !!(cpu
->env
.spr
[SPR_LPCR
] & LPCR_GTSE
);
1755 * Update LPCR:HR and UPRT as they may not be set properly in
1758 spapr_set_all_lpcrs(radix
? (LPCR_HR
| LPCR_UPRT
) : 0,
1759 LPCR_HR
| LPCR_UPRT
);
1761 err
= kvmppc_configure_v3_mmu(cpu
, radix
, gtse
, spapr
->patb_entry
);
1763 error_report("Process table config unsupported by the host");
1768 err
= spapr_irq_post_load(spapr
, version_id
);
1776 static int spapr_pre_save(void *opaque
)
1780 rc
= spapr_caps_pre_save(opaque
);
1788 static bool version_before_3(void *opaque
, int version_id
)
1790 return version_id
< 3;
1793 static bool spapr_pending_events_needed(void *opaque
)
1795 SpaprMachineState
*spapr
= (SpaprMachineState
*)opaque
;
1796 return !QTAILQ_EMPTY(&spapr
->pending_events
);
1799 static const VMStateDescription vmstate_spapr_event_entry
= {
1800 .name
= "spapr_event_log_entry",
1802 .minimum_version_id
= 1,
1803 .fields
= (VMStateField
[]) {
1804 VMSTATE_UINT32(summary
, SpaprEventLogEntry
),
1805 VMSTATE_UINT32(extended_length
, SpaprEventLogEntry
),
1806 VMSTATE_VBUFFER_ALLOC_UINT32(extended_log
, SpaprEventLogEntry
, 0,
1807 NULL
, extended_length
),
1808 VMSTATE_END_OF_LIST()
1812 static const VMStateDescription vmstate_spapr_pending_events
= {
1813 .name
= "spapr_pending_events",
1815 .minimum_version_id
= 1,
1816 .needed
= spapr_pending_events_needed
,
1817 .fields
= (VMStateField
[]) {
1818 VMSTATE_QTAILQ_V(pending_events
, SpaprMachineState
, 1,
1819 vmstate_spapr_event_entry
, SpaprEventLogEntry
, next
),
1820 VMSTATE_END_OF_LIST()
1824 static bool spapr_ov5_cas_needed(void *opaque
)
1826 SpaprMachineState
*spapr
= opaque
;
1827 SpaprOptionVector
*ov5_mask
= spapr_ovec_new();
1830 /* Prior to the introduction of SpaprOptionVector, we had two option
1831 * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1832 * Both of these options encode machine topology into the device-tree
1833 * in such a way that the now-booted OS should still be able to interact
1834 * appropriately with QEMU regardless of what options were actually
1835 * negotiatied on the source side.
1837 * As such, we can avoid migrating the CAS-negotiated options if these
1838 * are the only options available on the current machine/platform.
1839 * Since these are the only options available for pseries-2.7 and
1840 * earlier, this allows us to maintain old->new/new->old migration
1843 * For QEMU 2.8+, there are additional CAS-negotiatable options available
1844 * via default pseries-2.8 machines and explicit command-line parameters.
1845 * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1846 * of the actual CAS-negotiated values to continue working properly. For
1847 * example, availability of memory unplug depends on knowing whether
1848 * OV5_HP_EVT was negotiated via CAS.
1850 * Thus, for any cases where the set of available CAS-negotiatable
1851 * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1852 * include the CAS-negotiated options in the migration stream, unless
1853 * if they affect boot time behaviour only.
1855 spapr_ovec_set(ov5_mask
, OV5_FORM1_AFFINITY
);
1856 spapr_ovec_set(ov5_mask
, OV5_DRCONF_MEMORY
);
1857 spapr_ovec_set(ov5_mask
, OV5_DRMEM_V2
);
1859 /* We need extra information if we have any bits outside the mask
1861 cas_needed
= !spapr_ovec_subset(spapr
->ov5
, ov5_mask
);
1863 spapr_ovec_cleanup(ov5_mask
);
1868 static const VMStateDescription vmstate_spapr_ov5_cas
= {
1869 .name
= "spapr_option_vector_ov5_cas",
1871 .minimum_version_id
= 1,
1872 .needed
= spapr_ov5_cas_needed
,
1873 .fields
= (VMStateField
[]) {
1874 VMSTATE_STRUCT_POINTER_V(ov5_cas
, SpaprMachineState
, 1,
1875 vmstate_spapr_ovec
, SpaprOptionVector
),
1876 VMSTATE_END_OF_LIST()
1880 static bool spapr_patb_entry_needed(void *opaque
)
1882 SpaprMachineState
*spapr
= opaque
;
1884 return !!spapr
->patb_entry
;
1887 static const VMStateDescription vmstate_spapr_patb_entry
= {
1888 .name
= "spapr_patb_entry",
1890 .minimum_version_id
= 1,
1891 .needed
= spapr_patb_entry_needed
,
1892 .fields
= (VMStateField
[]) {
1893 VMSTATE_UINT64(patb_entry
, SpaprMachineState
),
1894 VMSTATE_END_OF_LIST()
1898 static bool spapr_irq_map_needed(void *opaque
)
1900 SpaprMachineState
*spapr
= opaque
;
1902 return spapr
->irq_map
&& !bitmap_empty(spapr
->irq_map
, spapr
->irq_map_nr
);
1905 static const VMStateDescription vmstate_spapr_irq_map
= {
1906 .name
= "spapr_irq_map",
1908 .minimum_version_id
= 1,
1909 .needed
= spapr_irq_map_needed
,
1910 .fields
= (VMStateField
[]) {
1911 VMSTATE_BITMAP(irq_map
, SpaprMachineState
, 0, irq_map_nr
),
1912 VMSTATE_END_OF_LIST()
1916 static bool spapr_dtb_needed(void *opaque
)
1918 SpaprMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(opaque
);
1920 return smc
->update_dt_enabled
;
1923 static int spapr_dtb_pre_load(void *opaque
)
1925 SpaprMachineState
*spapr
= (SpaprMachineState
*)opaque
;
1927 g_free(spapr
->fdt_blob
);
1928 spapr
->fdt_blob
= NULL
;
1929 spapr
->fdt_size
= 0;
1934 static const VMStateDescription vmstate_spapr_dtb
= {
1935 .name
= "spapr_dtb",
1937 .minimum_version_id
= 1,
1938 .needed
= spapr_dtb_needed
,
1939 .pre_load
= spapr_dtb_pre_load
,
1940 .fields
= (VMStateField
[]) {
1941 VMSTATE_UINT32(fdt_initial_size
, SpaprMachineState
),
1942 VMSTATE_UINT32(fdt_size
, SpaprMachineState
),
1943 VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob
, SpaprMachineState
, 0, NULL
,
1945 VMSTATE_END_OF_LIST()
1949 static bool spapr_fwnmi_needed(void *opaque
)
1951 SpaprMachineState
*spapr
= (SpaprMachineState
*)opaque
;
1953 return spapr
->fwnmi_machine_check_addr
!= -1;
1956 static int spapr_fwnmi_pre_save(void *opaque
)
1958 SpaprMachineState
*spapr
= (SpaprMachineState
*)opaque
;
1961 * Check if machine check handling is in progress and print a
1964 if (spapr
->fwnmi_machine_check_interlock
!= -1) {
1965 warn_report("A machine check is being handled during migration. The"
1966 "handler may run and log hardware error on the destination");
1972 static const VMStateDescription vmstate_spapr_fwnmi
= {
1973 .name
= "spapr_fwnmi",
1975 .minimum_version_id
= 1,
1976 .needed
= spapr_fwnmi_needed
,
1977 .pre_save
= spapr_fwnmi_pre_save
,
1978 .fields
= (VMStateField
[]) {
1979 VMSTATE_UINT64(fwnmi_system_reset_addr
, SpaprMachineState
),
1980 VMSTATE_UINT64(fwnmi_machine_check_addr
, SpaprMachineState
),
1981 VMSTATE_INT32(fwnmi_machine_check_interlock
, SpaprMachineState
),
1982 VMSTATE_END_OF_LIST()
1986 static const VMStateDescription vmstate_spapr
= {
1989 .minimum_version_id
= 1,
1990 .pre_load
= spapr_pre_load
,
1991 .post_load
= spapr_post_load
,
1992 .pre_save
= spapr_pre_save
,
1993 .fields
= (VMStateField
[]) {
1994 /* used to be @next_irq */
1995 VMSTATE_UNUSED_BUFFER(version_before_3
, 0, 4),
1998 VMSTATE_UINT64_TEST(rtc_offset
, SpaprMachineState
, version_before_3
),
2000 VMSTATE_PPC_TIMEBASE_V(tb
, SpaprMachineState
, 2),
2001 VMSTATE_END_OF_LIST()
2003 .subsections
= (const VMStateDescription
*[]) {
2004 &vmstate_spapr_ov5_cas
,
2005 &vmstate_spapr_patb_entry
,
2006 &vmstate_spapr_pending_events
,
2007 &vmstate_spapr_cap_htm
,
2008 &vmstate_spapr_cap_vsx
,
2009 &vmstate_spapr_cap_dfp
,
2010 &vmstate_spapr_cap_cfpc
,
2011 &vmstate_spapr_cap_sbbc
,
2012 &vmstate_spapr_cap_ibs
,
2013 &vmstate_spapr_cap_hpt_maxpagesize
,
2014 &vmstate_spapr_irq_map
,
2015 &vmstate_spapr_cap_nested_kvm_hv
,
2017 &vmstate_spapr_cap_large_decr
,
2018 &vmstate_spapr_cap_ccf_assist
,
2019 &vmstate_spapr_cap_fwnmi
,
2020 &vmstate_spapr_fwnmi
,
2025 static int htab_save_setup(QEMUFile
*f
, void *opaque
)
2027 SpaprMachineState
*spapr
= opaque
;
2029 /* "Iteration" header */
2030 if (!spapr
->htab_shift
) {
2031 qemu_put_be32(f
, -1);
2033 qemu_put_be32(f
, spapr
->htab_shift
);
2037 spapr
->htab_save_index
= 0;
2038 spapr
->htab_first_pass
= true;
2040 if (spapr
->htab_shift
) {
2041 assert(kvm_enabled());
2049 static void htab_save_chunk(QEMUFile
*f
, SpaprMachineState
*spapr
,
2050 int chunkstart
, int n_valid
, int n_invalid
)
2052 qemu_put_be32(f
, chunkstart
);
2053 qemu_put_be16(f
, n_valid
);
2054 qemu_put_be16(f
, n_invalid
);
2055 qemu_put_buffer(f
, HPTE(spapr
->htab
, chunkstart
),
2056 HASH_PTE_SIZE_64
* n_valid
);
2059 static void htab_save_end_marker(QEMUFile
*f
)
2061 qemu_put_be32(f
, 0);
2062 qemu_put_be16(f
, 0);
2063 qemu_put_be16(f
, 0);
2066 static void htab_save_first_pass(QEMUFile
*f
, SpaprMachineState
*spapr
,
2069 bool has_timeout
= max_ns
!= -1;
2070 int htabslots
= HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
;
2071 int index
= spapr
->htab_save_index
;
2072 int64_t starttime
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
2074 assert(spapr
->htab_first_pass
);
2079 /* Consume invalid HPTEs */
2080 while ((index
< htabslots
)
2081 && !HPTE_VALID(HPTE(spapr
->htab
, index
))) {
2082 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
2086 /* Consume valid HPTEs */
2088 while ((index
< htabslots
) && (index
- chunkstart
< USHRT_MAX
)
2089 && HPTE_VALID(HPTE(spapr
->htab
, index
))) {
2090 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
2094 if (index
> chunkstart
) {
2095 int n_valid
= index
- chunkstart
;
2097 htab_save_chunk(f
, spapr
, chunkstart
, n_valid
, 0);
2100 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - starttime
) > max_ns
) {
2104 } while ((index
< htabslots
) && !qemu_file_rate_limit(f
));
2106 if (index
>= htabslots
) {
2107 assert(index
== htabslots
);
2109 spapr
->htab_first_pass
= false;
2111 spapr
->htab_save_index
= index
;
2114 static int htab_save_later_pass(QEMUFile
*f
, SpaprMachineState
*spapr
,
2117 bool final
= max_ns
< 0;
2118 int htabslots
= HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
;
2119 int examined
= 0, sent
= 0;
2120 int index
= spapr
->htab_save_index
;
2121 int64_t starttime
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
2123 assert(!spapr
->htab_first_pass
);
2126 int chunkstart
, invalidstart
;
2128 /* Consume non-dirty HPTEs */
2129 while ((index
< htabslots
)
2130 && !HPTE_DIRTY(HPTE(spapr
->htab
, index
))) {
2136 /* Consume valid dirty HPTEs */
2137 while ((index
< htabslots
) && (index
- chunkstart
< USHRT_MAX
)
2138 && HPTE_DIRTY(HPTE(spapr
->htab
, index
))
2139 && HPTE_VALID(HPTE(spapr
->htab
, index
))) {
2140 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
2145 invalidstart
= index
;
2146 /* Consume invalid dirty HPTEs */
2147 while ((index
< htabslots
) && (index
- invalidstart
< USHRT_MAX
)
2148 && HPTE_DIRTY(HPTE(spapr
->htab
, index
))
2149 && !HPTE_VALID(HPTE(spapr
->htab
, index
))) {
2150 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
2155 if (index
> chunkstart
) {
2156 int n_valid
= invalidstart
- chunkstart
;
2157 int n_invalid
= index
- invalidstart
;
2159 htab_save_chunk(f
, spapr
, chunkstart
, n_valid
, n_invalid
);
2160 sent
+= index
- chunkstart
;
2162 if (!final
&& (qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - starttime
) > max_ns
) {
2167 if (examined
>= htabslots
) {
2171 if (index
>= htabslots
) {
2172 assert(index
== htabslots
);
2175 } while ((examined
< htabslots
) && (!qemu_file_rate_limit(f
) || final
));
2177 if (index
>= htabslots
) {
2178 assert(index
== htabslots
);
2182 spapr
->htab_save_index
= index
;
2184 return (examined
>= htabslots
) && (sent
== 0) ? 1 : 0;
2187 #define MAX_ITERATION_NS 5000000 /* 5 ms */
2188 #define MAX_KVM_BUF_SIZE 2048
2190 static int htab_save_iterate(QEMUFile
*f
, void *opaque
)
2192 SpaprMachineState
*spapr
= opaque
;
2196 /* Iteration header */
2197 if (!spapr
->htab_shift
) {
2198 qemu_put_be32(f
, -1);
2201 qemu_put_be32(f
, 0);
2205 assert(kvm_enabled());
2207 fd
= get_htab_fd(spapr
);
2212 rc
= kvmppc_save_htab(f
, fd
, MAX_KVM_BUF_SIZE
, MAX_ITERATION_NS
);
2216 } else if (spapr
->htab_first_pass
) {
2217 htab_save_first_pass(f
, spapr
, MAX_ITERATION_NS
);
2219 rc
= htab_save_later_pass(f
, spapr
, MAX_ITERATION_NS
);
2222 htab_save_end_marker(f
);
2227 static int htab_save_complete(QEMUFile
*f
, void *opaque
)
2229 SpaprMachineState
*spapr
= opaque
;
2232 /* Iteration header */
2233 if (!spapr
->htab_shift
) {
2234 qemu_put_be32(f
, -1);
2237 qemu_put_be32(f
, 0);
2243 assert(kvm_enabled());
2245 fd
= get_htab_fd(spapr
);
2250 rc
= kvmppc_save_htab(f
, fd
, MAX_KVM_BUF_SIZE
, -1);
2255 if (spapr
->htab_first_pass
) {
2256 htab_save_first_pass(f
, spapr
, -1);
2258 htab_save_later_pass(f
, spapr
, -1);
2262 htab_save_end_marker(f
);
2267 static int htab_load(QEMUFile
*f
, void *opaque
, int version_id
)
2269 SpaprMachineState
*spapr
= opaque
;
2270 uint32_t section_hdr
;
2272 Error
*local_err
= NULL
;
2274 if (version_id
< 1 || version_id
> 1) {
2275 error_report("htab_load() bad version");
2279 section_hdr
= qemu_get_be32(f
);
2281 if (section_hdr
== -1) {
2282 spapr_free_hpt(spapr
);
2289 /* First section gives the htab size */
2290 ret
= spapr_reallocate_hpt(spapr
, section_hdr
, &local_err
);
2292 error_report_err(local_err
);
2299 assert(kvm_enabled());
2301 fd
= kvmppc_get_htab_fd(true, 0, &local_err
);
2303 error_report_err(local_err
);
2310 uint16_t n_valid
, n_invalid
;
2312 index
= qemu_get_be32(f
);
2313 n_valid
= qemu_get_be16(f
);
2314 n_invalid
= qemu_get_be16(f
);
2316 if ((index
== 0) && (n_valid
== 0) && (n_invalid
== 0)) {
2321 if ((index
+ n_valid
+ n_invalid
) >
2322 (HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
)) {
2323 /* Bad index in stream */
2325 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2326 index
, n_valid
, n_invalid
, spapr
->htab_shift
);
2332 qemu_get_buffer(f
, HPTE(spapr
->htab
, index
),
2333 HASH_PTE_SIZE_64
* n_valid
);
2336 memset(HPTE(spapr
->htab
, index
+ n_valid
), 0,
2337 HASH_PTE_SIZE_64
* n_invalid
);
2344 rc
= kvmppc_load_htab_chunk(f
, fd
, index
, n_valid
, n_invalid
,
2347 error_report_err(local_err
);
2361 static void htab_save_cleanup(void *opaque
)
2363 SpaprMachineState
*spapr
= opaque
;
2365 close_htab_fd(spapr
);
2368 static SaveVMHandlers savevm_htab_handlers
= {
2369 .save_setup
= htab_save_setup
,
2370 .save_live_iterate
= htab_save_iterate
,
2371 .save_live_complete_precopy
= htab_save_complete
,
2372 .save_cleanup
= htab_save_cleanup
,
2373 .load_state
= htab_load
,
2376 static void spapr_boot_set(void *opaque
, const char *boot_device
,
2379 MachineState
*machine
= MACHINE(opaque
);
2380 machine
->boot_order
= g_strdup(boot_device
);
2383 static void spapr_create_lmb_dr_connectors(SpaprMachineState
*spapr
)
2385 MachineState
*machine
= MACHINE(spapr
);
2386 uint64_t lmb_size
= SPAPR_MEMORY_BLOCK_SIZE
;
2387 uint32_t nr_lmbs
= (machine
->maxram_size
- machine
->ram_size
)/lmb_size
;
2390 for (i
= 0; i
< nr_lmbs
; i
++) {
2393 addr
= i
* lmb_size
+ machine
->device_memory
->base
;
2394 spapr_dr_connector_new(OBJECT(spapr
), TYPE_SPAPR_DRC_LMB
,
2400 * If RAM size, maxmem size and individual node mem sizes aren't aligned
2401 * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2402 * since we can't support such unaligned sizes with DRCONF_MEMORY.
2404 static void spapr_validate_node_memory(MachineState
*machine
, Error
**errp
)
2408 if (machine
->ram_size
% SPAPR_MEMORY_BLOCK_SIZE
) {
2409 error_setg(errp
, "Memory size 0x" RAM_ADDR_FMT
2410 " is not aligned to %" PRIu64
" MiB",
2412 SPAPR_MEMORY_BLOCK_SIZE
/ MiB
);
2416 if (machine
->maxram_size
% SPAPR_MEMORY_BLOCK_SIZE
) {
2417 error_setg(errp
, "Maximum memory size 0x" RAM_ADDR_FMT
2418 " is not aligned to %" PRIu64
" MiB",
2420 SPAPR_MEMORY_BLOCK_SIZE
/ MiB
);
2424 for (i
= 0; i
< machine
->numa_state
->num_nodes
; i
++) {
2425 if (machine
->numa_state
->nodes
[i
].node_mem
% SPAPR_MEMORY_BLOCK_SIZE
) {
2427 "Node %d memory size 0x%" PRIx64
2428 " is not aligned to %" PRIu64
" MiB",
2429 i
, machine
->numa_state
->nodes
[i
].node_mem
,
2430 SPAPR_MEMORY_BLOCK_SIZE
/ MiB
);
2436 /* find cpu slot in machine->possible_cpus by core_id */
2437 static CPUArchId
*spapr_find_cpu_slot(MachineState
*ms
, uint32_t id
, int *idx
)
2439 int index
= id
/ ms
->smp
.threads
;
2441 if (index
>= ms
->possible_cpus
->len
) {
2447 return &ms
->possible_cpus
->cpus
[index
];
2450 static void spapr_set_vsmt_mode(SpaprMachineState
*spapr
, Error
**errp
)
2452 MachineState
*ms
= MACHINE(spapr
);
2453 SpaprMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(spapr
);
2454 Error
*local_err
= NULL
;
2455 bool vsmt_user
= !!spapr
->vsmt
;
2456 int kvm_smt
= kvmppc_smt_threads();
2458 unsigned int smp_threads
= ms
->smp
.threads
;
2460 if (!kvm_enabled() && (smp_threads
> 1)) {
2461 error_setg(errp
, "TCG cannot support more than 1 thread/core "
2462 "on a pseries machine");
2465 if (!is_power_of_2(smp_threads
)) {
2466 error_setg(errp
, "Cannot support %d threads/core on a pseries "
2467 "machine because it must be a power of 2", smp_threads
);
2471 /* Detemine the VSMT mode to use: */
2473 if (spapr
->vsmt
< smp_threads
) {
2474 error_setg(errp
, "Cannot support VSMT mode %d"
2475 " because it must be >= threads/core (%d)",
2476 spapr
->vsmt
, smp_threads
);
2479 /* In this case, spapr->vsmt has been set by the command line */
2480 } else if (!smc
->smp_threads_vsmt
) {
2482 * Default VSMT value is tricky, because we need it to be as
2483 * consistent as possible (for migration), but this requires
2484 * changing it for at least some existing cases. We pick 8 as
2485 * the value that we'd get with KVM on POWER8, the
2486 * overwhelmingly common case in production systems.
2488 spapr
->vsmt
= MAX(8, smp_threads
);
2490 spapr
->vsmt
= smp_threads
;
2493 /* KVM: If necessary, set the SMT mode: */
2494 if (kvm_enabled() && (spapr
->vsmt
!= kvm_smt
)) {
2495 ret
= kvmppc_set_smt_threads(spapr
->vsmt
);
2497 /* Looks like KVM isn't able to change VSMT mode */
2498 error_setg(&local_err
,
2499 "Failed to set KVM's VSMT mode to %d (errno %d)",
2501 /* We can live with that if the default one is big enough
2502 * for the number of threads, and a submultiple of the one
2503 * we want. In this case we'll waste some vcpu ids, but
2504 * behaviour will be correct */
2505 if ((kvm_smt
>= smp_threads
) && ((spapr
->vsmt
% kvm_smt
) == 0)) {
2506 warn_report_err(local_err
);
2509 error_append_hint(&local_err
,
2510 "On PPC, a VM with %d threads/core"
2511 " on a host with %d threads/core"
2512 " requires the use of VSMT mode %d.\n",
2513 smp_threads
, kvm_smt
, spapr
->vsmt
);
2515 kvmppc_error_append_smt_possible_hint(&local_err
);
2516 error_propagate(errp
, local_err
);
2520 /* else TCG: nothing to do currently */
2523 static void spapr_init_cpus(SpaprMachineState
*spapr
)
2525 MachineState
*machine
= MACHINE(spapr
);
2526 MachineClass
*mc
= MACHINE_GET_CLASS(machine
);
2527 SpaprMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(machine
);
2528 const char *type
= spapr_get_cpu_core_type(machine
->cpu_type
);
2529 const CPUArchIdList
*possible_cpus
;
2530 unsigned int smp_cpus
= machine
->smp
.cpus
;
2531 unsigned int smp_threads
= machine
->smp
.threads
;
2532 unsigned int max_cpus
= machine
->smp
.max_cpus
;
2533 int boot_cores_nr
= smp_cpus
/ smp_threads
;
2536 possible_cpus
= mc
->possible_cpu_arch_ids(machine
);
2537 if (mc
->has_hotpluggable_cpus
) {
2538 if (smp_cpus
% smp_threads
) {
2539 error_report("smp_cpus (%u) must be multiple of threads (%u)",
2540 smp_cpus
, smp_threads
);
2543 if (max_cpus
% smp_threads
) {
2544 error_report("max_cpus (%u) must be multiple of threads (%u)",
2545 max_cpus
, smp_threads
);
2549 if (max_cpus
!= smp_cpus
) {
2550 error_report("This machine version does not support CPU hotplug");
2553 boot_cores_nr
= possible_cpus
->len
;
2556 if (smc
->pre_2_10_has_unused_icps
) {
2559 for (i
= 0; i
< spapr_max_server_number(spapr
); i
++) {
2560 /* Dummy entries get deregistered when real ICPState objects
2561 * are registered during CPU core hotplug.
2563 pre_2_10_vmstate_register_dummy_icp(i
);
2567 for (i
= 0; i
< possible_cpus
->len
; i
++) {
2568 int core_id
= i
* smp_threads
;
2570 if (mc
->has_hotpluggable_cpus
) {
2571 spapr_dr_connector_new(OBJECT(spapr
), TYPE_SPAPR_DRC_CPU
,
2572 spapr_vcpu_id(spapr
, core_id
));
2575 if (i
< boot_cores_nr
) {
2576 Object
*core
= object_new(type
);
2577 int nr_threads
= smp_threads
;
2579 /* Handle the partially filled core for older machine types */
2580 if ((i
+ 1) * smp_threads
>= smp_cpus
) {
2581 nr_threads
= smp_cpus
- i
* smp_threads
;
2584 object_property_set_int(core
, "nr-threads", nr_threads
,
2586 object_property_set_int(core
, CPU_CORE_PROP_CORE_ID
, core_id
,
2588 qdev_realize(DEVICE(core
), NULL
, &error_fatal
);
2595 static PCIHostState
*spapr_create_default_phb(void)
2599 dev
= qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE
);
2600 qdev_prop_set_uint32(dev
, "index", 0);
2601 sysbus_realize_and_unref(SYS_BUS_DEVICE(dev
), &error_fatal
);
2603 return PCI_HOST_BRIDGE(dev
);
2606 static hwaddr
spapr_rma_size(SpaprMachineState
*spapr
, Error
**errp
)
2608 MachineState
*machine
= MACHINE(spapr
);
2609 SpaprMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(spapr
);
2610 hwaddr rma_size
= machine
->ram_size
;
2611 hwaddr node0_size
= spapr_node0_size(machine
);
2613 /* RMA has to fit in the first NUMA node */
2614 rma_size
= MIN(rma_size
, node0_size
);
2617 * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2620 rma_size
= MIN(rma_size
, 1 * TiB
);
2623 * Clamp the RMA size based on machine type. This is for
2624 * migration compatibility with older qemu versions, which limited
2625 * the RMA size for complicated and mostly bad reasons.
2627 if (smc
->rma_limit
) {
2628 rma_size
= MIN(rma_size
, smc
->rma_limit
);
2631 if (rma_size
< MIN_RMA_SLOF
) {
2633 "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2634 "ldMiB guest RMA (Real Mode Area memory)",
2635 MIN_RMA_SLOF
/ MiB
);
2642 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState
*spapr
)
2644 MachineState
*machine
= MACHINE(spapr
);
2647 for (i
= 0; i
< machine
->ram_slots
; i
++) {
2648 spapr_dr_connector_new(OBJECT(spapr
), TYPE_SPAPR_DRC_PMEM
, i
);
2652 /* pSeries LPAR / sPAPR hardware init */
2653 static void spapr_machine_init(MachineState
*machine
)
2655 SpaprMachineState
*spapr
= SPAPR_MACHINE(machine
);
2656 SpaprMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(machine
);
2657 MachineClass
*mc
= MACHINE_GET_CLASS(machine
);
2658 const char *bios_name
= machine
->firmware
?: FW_FILE_NAME
;
2659 const char *kernel_filename
= machine
->kernel_filename
;
2660 const char *initrd_filename
= machine
->initrd_filename
;
2663 MemoryRegion
*sysmem
= get_system_memory();
2664 long load_limit
, fw_size
;
2666 Error
*resize_hpt_err
= NULL
;
2669 * if Secure VM (PEF) support is configured, then initialize it
2671 pef_kvm_init(machine
->cgs
, &error_fatal
);
2673 msi_nonbroken
= true;
2675 QLIST_INIT(&spapr
->phbs
);
2676 QTAILQ_INIT(&spapr
->pending_dimm_unplugs
);
2678 /* Determine capabilities to run with */
2679 spapr_caps_init(spapr
);
2681 kvmppc_check_papr_resize_hpt(&resize_hpt_err
);
2682 if (spapr
->resize_hpt
== SPAPR_RESIZE_HPT_DEFAULT
) {
2684 * If the user explicitly requested a mode we should either
2685 * supply it, or fail completely (which we do below). But if
2686 * it's not set explicitly, we reset our mode to something
2689 if (resize_hpt_err
) {
2690 spapr
->resize_hpt
= SPAPR_RESIZE_HPT_DISABLED
;
2691 error_free(resize_hpt_err
);
2692 resize_hpt_err
= NULL
;
2694 spapr
->resize_hpt
= smc
->resize_hpt_default
;
2698 assert(spapr
->resize_hpt
!= SPAPR_RESIZE_HPT_DEFAULT
);
2700 if ((spapr
->resize_hpt
!= SPAPR_RESIZE_HPT_DISABLED
) && resize_hpt_err
) {
2702 * User requested HPT resize, but this host can't supply it. Bail out
2704 error_report_err(resize_hpt_err
);
2707 error_free(resize_hpt_err
);
2709 spapr
->rma_size
= spapr_rma_size(spapr
, &error_fatal
);
2711 /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2712 load_limit
= MIN(spapr
->rma_size
, FDT_MAX_ADDR
) - FW_OVERHEAD
;
2715 * VSMT must be set in order to be able to compute VCPU ids, ie to
2716 * call spapr_max_server_number() or spapr_vcpu_id().
2718 spapr_set_vsmt_mode(spapr
, &error_fatal
);
2720 /* Set up Interrupt Controller before we create the VCPUs */
2721 spapr_irq_init(spapr
, &error_fatal
);
2723 /* Set up containers for ibm,client-architecture-support negotiated options
2725 spapr
->ov5
= spapr_ovec_new();
2726 spapr
->ov5_cas
= spapr_ovec_new();
2728 if (smc
->dr_lmb_enabled
) {
2729 spapr_ovec_set(spapr
->ov5
, OV5_DRCONF_MEMORY
);
2730 spapr_validate_node_memory(machine
, &error_fatal
);
2733 spapr_ovec_set(spapr
->ov5
, OV5_FORM1_AFFINITY
);
2735 /* advertise support for dedicated HP event source to guests */
2736 if (spapr
->use_hotplug_event_source
) {
2737 spapr_ovec_set(spapr
->ov5
, OV5_HP_EVT
);
2740 /* advertise support for HPT resizing */
2741 if (spapr
->resize_hpt
!= SPAPR_RESIZE_HPT_DISABLED
) {
2742 spapr_ovec_set(spapr
->ov5
, OV5_HPT_RESIZE
);
2745 /* advertise support for ibm,dyamic-memory-v2 */
2746 spapr_ovec_set(spapr
->ov5
, OV5_DRMEM_V2
);
2748 /* advertise XIVE on POWER9 machines */
2749 if (spapr
->irq
->xive
) {
2750 spapr_ovec_set(spapr
->ov5
, OV5_XIVE_EXPLOIT
);
2754 spapr_init_cpus(spapr
);
2757 * check we don't have a memory-less/cpu-less NUMA node
2758 * Firmware relies on the existing memory/cpu topology to provide the
2759 * NUMA topology to the kernel.
2760 * And the linux kernel needs to know the NUMA topology at start
2761 * to be able to hotplug CPUs later.
2763 if (machine
->numa_state
->num_nodes
) {
2764 for (i
= 0; i
< machine
->numa_state
->num_nodes
; ++i
) {
2765 /* check for memory-less node */
2766 if (machine
->numa_state
->nodes
[i
].node_mem
== 0) {
2769 /* check for cpu-less node */
2771 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
2772 if (cpu
->node_id
== i
) {
2777 /* memory-less and cpu-less node */
2780 "Memory-less/cpu-less nodes are not supported (node %d)",
2789 spapr
->gpu_numa_id
= spapr_numa_initial_nvgpu_numa_id(machine
);
2791 /* Init numa_assoc_array */
2792 spapr_numa_associativity_init(spapr
, machine
);
2794 if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2795 ppc_type_check_compat(machine
->cpu_type
, CPU_POWERPC_LOGICAL_3_00
, 0,
2796 spapr
->max_compat_pvr
)) {
2797 spapr_ovec_set(spapr
->ov5
, OV5_MMU_RADIX_300
);
2798 /* KVM and TCG always allow GTSE with radix... */
2799 spapr_ovec_set(spapr
->ov5
, OV5_MMU_RADIX_GTSE
);
2801 /* ... but not with hash (currently). */
2803 if (kvm_enabled()) {
2804 /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2805 kvmppc_enable_logical_ci_hcalls();
2806 kvmppc_enable_set_mode_hcall();
2808 /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2809 kvmppc_enable_clear_ref_mod_hcalls();
2811 /* Enable H_PAGE_INIT */
2812 kvmppc_enable_h_page_init();
2816 memory_region_add_subregion(sysmem
, 0, machine
->ram
);
2818 /* always allocate the device memory information */
2819 machine
->device_memory
= g_malloc0(sizeof(*machine
->device_memory
));
2821 /* initialize hotplug memory address space */
2822 if (machine
->ram_size
< machine
->maxram_size
) {
2823 ram_addr_t device_mem_size
= machine
->maxram_size
- machine
->ram_size
;
2825 * Limit the number of hotpluggable memory slots to half the number
2826 * slots that KVM supports, leaving the other half for PCI and other
2827 * devices. However ensure that number of slots doesn't drop below 32.
2829 int max_memslots
= kvm_enabled() ? kvm_get_max_memslots() / 2 :
2830 SPAPR_MAX_RAM_SLOTS
;
2832 if (max_memslots
< SPAPR_MAX_RAM_SLOTS
) {
2833 max_memslots
= SPAPR_MAX_RAM_SLOTS
;
2835 if (machine
->ram_slots
> max_memslots
) {
2836 error_report("Specified number of memory slots %"
2837 PRIu64
" exceeds max supported %d",
2838 machine
->ram_slots
, max_memslots
);
2842 machine
->device_memory
->base
= ROUND_UP(machine
->ram_size
,
2843 SPAPR_DEVICE_MEM_ALIGN
);
2844 memory_region_init(&machine
->device_memory
->mr
, OBJECT(spapr
),
2845 "device-memory", device_mem_size
);
2846 memory_region_add_subregion(sysmem
, machine
->device_memory
->base
,
2847 &machine
->device_memory
->mr
);
2850 if (smc
->dr_lmb_enabled
) {
2851 spapr_create_lmb_dr_connectors(spapr
);
2854 if (spapr_get_cap(spapr
, SPAPR_CAP_FWNMI
) == SPAPR_CAP_ON
) {
2855 /* Create the error string for live migration blocker */
2856 error_setg(&spapr
->fwnmi_migration_blocker
,
2857 "A machine check is being handled during migration. The handler"
2858 "may run and log hardware error on the destination");
2861 if (mc
->nvdimm_supported
) {
2862 spapr_create_nvdimm_dr_connectors(spapr
);
2865 /* Set up RTAS event infrastructure */
2866 spapr_events_init(spapr
);
2868 /* Set up the RTC RTAS interfaces */
2869 spapr_rtc_create(spapr
);
2871 /* Set up VIO bus */
2872 spapr
->vio_bus
= spapr_vio_bus_init();
2874 for (i
= 0; serial_hd(i
); i
++) {
2875 spapr_vty_create(spapr
->vio_bus
, serial_hd(i
));
2878 /* We always have at least the nvram device on VIO */
2879 spapr_create_nvram(spapr
);
2882 * Setup hotplug / dynamic-reconfiguration connectors. top-level
2883 * connectors (described in root DT node's "ibm,drc-types" property)
2884 * are pre-initialized here. additional child connectors (such as
2885 * connectors for a PHBs PCI slots) are added as needed during their
2886 * parent's realization.
2888 if (smc
->dr_phb_enabled
) {
2889 for (i
= 0; i
< SPAPR_MAX_PHBS
; i
++) {
2890 spapr_dr_connector_new(OBJECT(machine
), TYPE_SPAPR_DRC_PHB
, i
);
2895 spapr_pci_rtas_init();
2897 phb
= spapr_create_default_phb();
2899 for (i
= 0; i
< nb_nics
; i
++) {
2900 NICInfo
*nd
= &nd_table
[i
];
2903 nd
->model
= g_strdup("spapr-vlan");
2906 if (g_str_equal(nd
->model
, "spapr-vlan") ||
2907 g_str_equal(nd
->model
, "ibmveth")) {
2908 spapr_vlan_create(spapr
->vio_bus
, nd
);
2910 pci_nic_init_nofail(&nd_table
[i
], phb
->bus
, nd
->model
, NULL
);
2914 for (i
= 0; i
<= drive_get_max_bus(IF_SCSI
); i
++) {
2915 spapr_vscsi_create(spapr
->vio_bus
);
2919 if (spapr_vga_init(phb
->bus
, &error_fatal
)) {
2920 spapr
->has_graphics
= true;
2921 machine
->usb
|= defaults_enabled() && !machine
->usb_disabled
;
2925 if (smc
->use_ohci_by_default
) {
2926 pci_create_simple(phb
->bus
, -1, "pci-ohci");
2928 pci_create_simple(phb
->bus
, -1, "nec-usb-xhci");
2931 if (spapr
->has_graphics
) {
2932 USBBus
*usb_bus
= usb_bus_find(-1);
2934 usb_create_simple(usb_bus
, "usb-kbd");
2935 usb_create_simple(usb_bus
, "usb-mouse");
2939 if (kernel_filename
) {
2940 spapr
->kernel_size
= load_elf(kernel_filename
, NULL
,
2941 translate_kernel_address
, spapr
,
2942 NULL
, NULL
, NULL
, NULL
, 1,
2943 PPC_ELF_MACHINE
, 0, 0);
2944 if (spapr
->kernel_size
== ELF_LOAD_WRONG_ENDIAN
) {
2945 spapr
->kernel_size
= load_elf(kernel_filename
, NULL
,
2946 translate_kernel_address
, spapr
,
2947 NULL
, NULL
, NULL
, NULL
, 0,
2948 PPC_ELF_MACHINE
, 0, 0);
2949 spapr
->kernel_le
= spapr
->kernel_size
> 0;
2951 if (spapr
->kernel_size
< 0) {
2952 error_report("error loading %s: %s", kernel_filename
,
2953 load_elf_strerror(spapr
->kernel_size
));
2958 if (initrd_filename
) {
2959 /* Try to locate the initrd in the gap between the kernel
2960 * and the firmware. Add a bit of space just in case
2962 spapr
->initrd_base
= (spapr
->kernel_addr
+ spapr
->kernel_size
2963 + 0x1ffff) & ~0xffff;
2964 spapr
->initrd_size
= load_image_targphys(initrd_filename
,
2967 - spapr
->initrd_base
);
2968 if (spapr
->initrd_size
< 0) {
2969 error_report("could not load initial ram disk '%s'",
2976 filename
= qemu_find_file(QEMU_FILE_TYPE_BIOS
, bios_name
);
2978 error_report("Could not find LPAR firmware '%s'", bios_name
);
2981 fw_size
= load_image_targphys(filename
, 0, FW_MAX_SIZE
);
2983 error_report("Could not load LPAR firmware '%s'", filename
);
2988 /* FIXME: Should register things through the MachineState's qdev
2989 * interface, this is a legacy from the sPAPREnvironment structure
2990 * which predated MachineState but had a similar function */
2991 vmstate_register(NULL
, 0, &vmstate_spapr
, spapr
);
2992 register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY
, 1,
2993 &savevm_htab_handlers
, spapr
);
2995 qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine
));
2997 qemu_register_boot_set(spapr_boot_set
, spapr
);
3000 * Nothing needs to be done to resume a suspended guest because
3001 * suspending does not change the machine state, so no need for
3002 * a ->wakeup method.
3004 qemu_register_wakeup_support();
3006 if (kvm_enabled()) {
3007 /* to stop and start vmclock */
3008 qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change
,
3011 kvmppc_spapr_enable_inkernel_multitce();
3014 qemu_cond_init(&spapr
->fwnmi_machine_check_interlock_cond
);
3017 #define DEFAULT_KVM_TYPE "auto"
3018 static int spapr_kvm_type(MachineState
*machine
, const char *vm_type
)
3021 * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3022 * accomodate the 'HV' and 'PV' formats that exists in the
3023 * wild. The 'auto' mode is being introduced already as
3024 * lower-case, thus we don't need to bother checking for
3027 if (!vm_type
|| !strcmp(vm_type
, DEFAULT_KVM_TYPE
)) {
3031 if (!g_ascii_strcasecmp(vm_type
, "hv")) {
3035 if (!g_ascii_strcasecmp(vm_type
, "pr")) {
3039 error_report("Unknown kvm-type specified '%s'", vm_type
);
3044 * Implementation of an interface to adjust firmware path
3045 * for the bootindex property handling.
3047 static char *spapr_get_fw_dev_path(FWPathProvider
*p
, BusState
*bus
,
3050 #define CAST(type, obj, name) \
3051 ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3052 SCSIDevice
*d
= CAST(SCSIDevice
, dev
, TYPE_SCSI_DEVICE
);
3053 SpaprPhbState
*phb
= CAST(SpaprPhbState
, dev
, TYPE_SPAPR_PCI_HOST_BRIDGE
);
3054 VHostSCSICommon
*vsc
= CAST(VHostSCSICommon
, dev
, TYPE_VHOST_SCSI_COMMON
);
3055 PCIDevice
*pcidev
= CAST(PCIDevice
, dev
, TYPE_PCI_DEVICE
);
3058 void *spapr
= CAST(void, bus
->parent
, "spapr-vscsi");
3059 VirtIOSCSI
*virtio
= CAST(VirtIOSCSI
, bus
->parent
, TYPE_VIRTIO_SCSI
);
3060 USBDevice
*usb
= CAST(USBDevice
, bus
->parent
, TYPE_USB_DEVICE
);
3064 * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3065 * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3066 * 0x8000 | (target << 8) | (bus << 5) | lun
3067 * (see the "Logical unit addressing format" table in SAM5)
3069 unsigned id
= 0x8000 | (d
->id
<< 8) | (d
->channel
<< 5) | d
->lun
;
3070 return g_strdup_printf("%s@%"PRIX64
, qdev_fw_name(dev
),
3071 (uint64_t)id
<< 48);
3072 } else if (virtio
) {
3074 * We use SRP luns of the form 01000000 | (target << 8) | lun
3075 * in the top 32 bits of the 64-bit LUN
3076 * Note: the quote above is from SLOF and it is wrong,
3077 * the actual binding is:
3078 * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3080 unsigned id
= 0x1000000 | (d
->id
<< 16) | d
->lun
;
3081 if (d
->lun
>= 256) {
3082 /* Use the LUN "flat space addressing method" */
3085 return g_strdup_printf("%s@%"PRIX64
, qdev_fw_name(dev
),
3086 (uint64_t)id
<< 32);
3089 * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3090 * in the top 32 bits of the 64-bit LUN
3092 unsigned usb_port
= atoi(usb
->port
->path
);
3093 unsigned id
= 0x1000000 | (usb_port
<< 16) | d
->lun
;
3094 return g_strdup_printf("%s@%"PRIX64
, qdev_fw_name(dev
),
3095 (uint64_t)id
<< 32);
3100 * SLOF probes the USB devices, and if it recognizes that the device is a
3101 * storage device, it changes its name to "storage" instead of "usb-host",
3102 * and additionally adds a child node for the SCSI LUN, so the correct
3103 * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3105 if (strcmp("usb-host", qdev_fw_name(dev
)) == 0) {
3106 USBDevice
*usbdev
= CAST(USBDevice
, dev
, TYPE_USB_DEVICE
);
3107 if (usb_host_dev_is_scsi_storage(usbdev
)) {
3108 return g_strdup_printf("storage@%s/disk", usbdev
->port
->path
);
3113 /* Replace "pci" with "pci@800000020000000" */
3114 return g_strdup_printf("pci@%"PRIX64
, phb
->buid
);
3118 /* Same logic as virtio above */
3119 unsigned id
= 0x1000000 | (vsc
->target
<< 16) | vsc
->lun
;
3120 return g_strdup_printf("disk@%"PRIX64
, (uint64_t)id
<< 32);
3123 if (g_str_equal("pci-bridge", qdev_fw_name(dev
))) {
3124 /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3125 PCIDevice
*pcidev
= CAST(PCIDevice
, dev
, TYPE_PCI_DEVICE
);
3126 return g_strdup_printf("pci@%x", PCI_SLOT(pcidev
->devfn
));
3130 return spapr_pci_fw_dev_name(pcidev
);
3136 static char *spapr_get_kvm_type(Object
*obj
, Error
**errp
)
3138 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
3140 return g_strdup(spapr
->kvm_type
);
3143 static void spapr_set_kvm_type(Object
*obj
, const char *value
, Error
**errp
)
3145 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
3147 g_free(spapr
->kvm_type
);
3148 spapr
->kvm_type
= g_strdup(value
);
3151 static bool spapr_get_modern_hotplug_events(Object
*obj
, Error
**errp
)
3153 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
3155 return spapr
->use_hotplug_event_source
;
3158 static void spapr_set_modern_hotplug_events(Object
*obj
, bool value
,
3161 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
3163 spapr
->use_hotplug_event_source
= value
;
3166 static bool spapr_get_msix_emulation(Object
*obj
, Error
**errp
)
3171 static char *spapr_get_resize_hpt(Object
*obj
, Error
**errp
)
3173 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
3175 switch (spapr
->resize_hpt
) {
3176 case SPAPR_RESIZE_HPT_DEFAULT
:
3177 return g_strdup("default");
3178 case SPAPR_RESIZE_HPT_DISABLED
:
3179 return g_strdup("disabled");
3180 case SPAPR_RESIZE_HPT_ENABLED
:
3181 return g_strdup("enabled");
3182 case SPAPR_RESIZE_HPT_REQUIRED
:
3183 return g_strdup("required");
3185 g_assert_not_reached();
3188 static void spapr_set_resize_hpt(Object
*obj
, const char *value
, Error
**errp
)
3190 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
3192 if (strcmp(value
, "default") == 0) {
3193 spapr
->resize_hpt
= SPAPR_RESIZE_HPT_DEFAULT
;
3194 } else if (strcmp(value
, "disabled") == 0) {
3195 spapr
->resize_hpt
= SPAPR_RESIZE_HPT_DISABLED
;
3196 } else if (strcmp(value
, "enabled") == 0) {
3197 spapr
->resize_hpt
= SPAPR_RESIZE_HPT_ENABLED
;
3198 } else if (strcmp(value
, "required") == 0) {
3199 spapr
->resize_hpt
= SPAPR_RESIZE_HPT_REQUIRED
;
3201 error_setg(errp
, "Bad value for \"resize-hpt\" property");
3205 static char *spapr_get_ic_mode(Object
*obj
, Error
**errp
)
3207 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
3209 if (spapr
->irq
== &spapr_irq_xics_legacy
) {
3210 return g_strdup("legacy");
3211 } else if (spapr
->irq
== &spapr_irq_xics
) {
3212 return g_strdup("xics");
3213 } else if (spapr
->irq
== &spapr_irq_xive
) {
3214 return g_strdup("xive");
3215 } else if (spapr
->irq
== &spapr_irq_dual
) {
3216 return g_strdup("dual");
3218 g_assert_not_reached();
3221 static void spapr_set_ic_mode(Object
*obj
, const char *value
, Error
**errp
)
3223 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
3225 if (SPAPR_MACHINE_GET_CLASS(spapr
)->legacy_irq_allocation
) {
3226 error_setg(errp
, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3230 /* The legacy IRQ backend can not be set */
3231 if (strcmp(value
, "xics") == 0) {
3232 spapr
->irq
= &spapr_irq_xics
;
3233 } else if (strcmp(value
, "xive") == 0) {
3234 spapr
->irq
= &spapr_irq_xive
;
3235 } else if (strcmp(value
, "dual") == 0) {
3236 spapr
->irq
= &spapr_irq_dual
;
3238 error_setg(errp
, "Bad value for \"ic-mode\" property");
3242 static char *spapr_get_host_model(Object
*obj
, Error
**errp
)
3244 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
3246 return g_strdup(spapr
->host_model
);
3249 static void spapr_set_host_model(Object
*obj
, const char *value
, Error
**errp
)
3251 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
3253 g_free(spapr
->host_model
);
3254 spapr
->host_model
= g_strdup(value
);
3257 static char *spapr_get_host_serial(Object
*obj
, Error
**errp
)
3259 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
3261 return g_strdup(spapr
->host_serial
);
3264 static void spapr_set_host_serial(Object
*obj
, const char *value
, Error
**errp
)
3266 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
3268 g_free(spapr
->host_serial
);
3269 spapr
->host_serial
= g_strdup(value
);
3272 static void spapr_instance_init(Object
*obj
)
3274 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
3275 SpaprMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(spapr
);
3276 MachineState
*ms
= MACHINE(spapr
);
3277 MachineClass
*mc
= MACHINE_GET_CLASS(ms
);
3280 * NVDIMM support went live in 5.1 without considering that, in
3281 * other archs, the user needs to enable NVDIMM support with the
3282 * 'nvdimm' machine option and the default behavior is NVDIMM
3283 * support disabled. It is too late to roll back to the standard
3284 * behavior without breaking 5.1 guests.
3286 if (mc
->nvdimm_supported
) {
3287 ms
->nvdimms_state
->is_enabled
= true;
3290 spapr
->htab_fd
= -1;
3291 spapr
->use_hotplug_event_source
= true;
3292 spapr
->kvm_type
= g_strdup(DEFAULT_KVM_TYPE
);
3293 object_property_add_str(obj
, "kvm-type",
3294 spapr_get_kvm_type
, spapr_set_kvm_type
);
3295 object_property_set_description(obj
, "kvm-type",
3296 "Specifies the KVM virtualization mode (auto,"
3297 " hv, pr). Defaults to 'auto'. This mode will use"
3298 " any available KVM module loaded in the host,"
3299 " where kvm_hv takes precedence if both kvm_hv and"
3300 " kvm_pr are loaded.");
3301 object_property_add_bool(obj
, "modern-hotplug-events",
3302 spapr_get_modern_hotplug_events
,
3303 spapr_set_modern_hotplug_events
);
3304 object_property_set_description(obj
, "modern-hotplug-events",
3305 "Use dedicated hotplug event mechanism in"
3306 " place of standard EPOW events when possible"
3307 " (required for memory hot-unplug support)");
3308 ppc_compat_add_property(obj
, "max-cpu-compat", &spapr
->max_compat_pvr
,
3309 "Maximum permitted CPU compatibility mode");
3311 object_property_add_str(obj
, "resize-hpt",
3312 spapr_get_resize_hpt
, spapr_set_resize_hpt
);
3313 object_property_set_description(obj
, "resize-hpt",
3314 "Resizing of the Hash Page Table (enabled, disabled, required)");
3315 object_property_add_uint32_ptr(obj
, "vsmt",
3316 &spapr
->vsmt
, OBJ_PROP_FLAG_READWRITE
);
3317 object_property_set_description(obj
, "vsmt",
3318 "Virtual SMT: KVM behaves as if this were"
3319 " the host's SMT mode");
3321 object_property_add_bool(obj
, "vfio-no-msix-emulation",
3322 spapr_get_msix_emulation
, NULL
);
3324 object_property_add_uint64_ptr(obj
, "kernel-addr",
3325 &spapr
->kernel_addr
, OBJ_PROP_FLAG_READWRITE
);
3326 object_property_set_description(obj
, "kernel-addr",
3327 stringify(KERNEL_LOAD_ADDR
)
3328 " for -kernel is the default");
3329 spapr
->kernel_addr
= KERNEL_LOAD_ADDR
;
3330 /* The machine class defines the default interrupt controller mode */
3331 spapr
->irq
= smc
->irq
;
3332 object_property_add_str(obj
, "ic-mode", spapr_get_ic_mode
,
3334 object_property_set_description(obj
, "ic-mode",
3335 "Specifies the interrupt controller mode (xics, xive, dual)");
3337 object_property_add_str(obj
, "host-model",
3338 spapr_get_host_model
, spapr_set_host_model
);
3339 object_property_set_description(obj
, "host-model",
3340 "Host model to advertise in guest device tree");
3341 object_property_add_str(obj
, "host-serial",
3342 spapr_get_host_serial
, spapr_set_host_serial
);
3343 object_property_set_description(obj
, "host-serial",
3344 "Host serial number to advertise in guest device tree");
3347 static void spapr_machine_finalizefn(Object
*obj
)
3349 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
3351 g_free(spapr
->kvm_type
);
3354 void spapr_do_system_reset_on_cpu(CPUState
*cs
, run_on_cpu_data arg
)
3356 SpaprMachineState
*spapr
= SPAPR_MACHINE(qdev_get_machine());
3357 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
3358 CPUPPCState
*env
= &cpu
->env
;
3360 cpu_synchronize_state(cs
);
3361 /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3362 if (spapr
->fwnmi_system_reset_addr
!= -1) {
3363 uint64_t rtas_addr
, addr
;
3365 /* get rtas addr from fdt */
3366 rtas_addr
= spapr_get_rtas_addr();
3368 qemu_system_guest_panicked(NULL
);
3372 addr
= rtas_addr
+ RTAS_ERROR_LOG_MAX
+ cs
->cpu_index
* sizeof(uint64_t)*2;
3373 stq_be_phys(&address_space_memory
, addr
, env
->gpr
[3]);
3374 stq_be_phys(&address_space_memory
, addr
+ sizeof(uint64_t), 0);
3377 ppc_cpu_do_system_reset(cs
);
3378 if (spapr
->fwnmi_system_reset_addr
!= -1) {
3379 env
->nip
= spapr
->fwnmi_system_reset_addr
;
3383 static void spapr_nmi(NMIState
*n
, int cpu_index
, Error
**errp
)
3388 async_run_on_cpu(cs
, spapr_do_system_reset_on_cpu
, RUN_ON_CPU_NULL
);
3392 int spapr_lmb_dt_populate(SpaprDrc
*drc
, SpaprMachineState
*spapr
,
3393 void *fdt
, int *fdt_start_offset
, Error
**errp
)
3398 addr
= spapr_drc_index(drc
) * SPAPR_MEMORY_BLOCK_SIZE
;
3399 node
= object_property_get_uint(OBJECT(drc
->dev
), PC_DIMM_NODE_PROP
,
3401 *fdt_start_offset
= spapr_dt_memory_node(spapr
, fdt
, node
, addr
,
3402 SPAPR_MEMORY_BLOCK_SIZE
);
3406 static void spapr_add_lmbs(DeviceState
*dev
, uint64_t addr_start
, uint64_t size
,
3407 bool dedicated_hp_event_source
)
3410 uint32_t nr_lmbs
= size
/SPAPR_MEMORY_BLOCK_SIZE
;
3412 uint64_t addr
= addr_start
;
3413 bool hotplugged
= spapr_drc_hotplugged(dev
);
3415 for (i
= 0; i
< nr_lmbs
; i
++) {
3416 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
3417 addr
/ SPAPR_MEMORY_BLOCK_SIZE
);
3421 * memory_device_get_free_addr() provided a range of free addresses
3422 * that doesn't overlap with any existing mapping at pre-plug. The
3423 * corresponding LMB DRCs are thus assumed to be all attachable.
3425 spapr_drc_attach(drc
, dev
);
3427 spapr_drc_reset(drc
);
3429 addr
+= SPAPR_MEMORY_BLOCK_SIZE
;
3431 /* send hotplug notification to the
3432 * guest only in case of hotplugged memory
3435 if (dedicated_hp_event_source
) {
3436 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
3437 addr_start
/ SPAPR_MEMORY_BLOCK_SIZE
);
3439 spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB
,
3441 spapr_drc_index(drc
));
3443 spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB
,
3449 static void spapr_memory_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
)
3451 SpaprMachineState
*ms
= SPAPR_MACHINE(hotplug_dev
);
3452 PCDIMMDevice
*dimm
= PC_DIMM(dev
);
3453 uint64_t size
, addr
;
3455 bool is_nvdimm
= object_dynamic_cast(OBJECT(dev
), TYPE_NVDIMM
);
3457 size
= memory_device_get_region_size(MEMORY_DEVICE(dev
), &error_abort
);
3459 pc_dimm_plug(dimm
, MACHINE(ms
));
3462 addr
= object_property_get_uint(OBJECT(dimm
),
3463 PC_DIMM_ADDR_PROP
, &error_abort
);
3464 spapr_add_lmbs(dev
, addr
, size
,
3465 spapr_ovec_test(ms
->ov5_cas
, OV5_HP_EVT
));
3467 slot
= object_property_get_int(OBJECT(dimm
),
3468 PC_DIMM_SLOT_PROP
, &error_abort
);
3469 /* We should have valid slot number at this point */
3470 g_assert(slot
>= 0);
3471 spapr_add_nvdimm(dev
, slot
);
3475 static void spapr_memory_pre_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
3478 const SpaprMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(hotplug_dev
);
3479 SpaprMachineState
*spapr
= SPAPR_MACHINE(hotplug_dev
);
3480 bool is_nvdimm
= object_dynamic_cast(OBJECT(dev
), TYPE_NVDIMM
);
3481 PCDIMMDevice
*dimm
= PC_DIMM(dev
);
3482 Error
*local_err
= NULL
;
3487 if (!smc
->dr_lmb_enabled
) {
3488 error_setg(errp
, "Memory hotplug not supported for this machine");
3492 size
= memory_device_get_region_size(MEMORY_DEVICE(dimm
), &local_err
);
3494 error_propagate(errp
, local_err
);
3499 if (!spapr_nvdimm_validate(hotplug_dev
, NVDIMM(dev
), size
, errp
)) {
3502 } else if (size
% SPAPR_MEMORY_BLOCK_SIZE
) {
3503 error_setg(errp
, "Hotplugged memory size must be a multiple of "
3504 "%" PRIu64
" MB", SPAPR_MEMORY_BLOCK_SIZE
/ MiB
);
3508 memdev
= object_property_get_link(OBJECT(dimm
), PC_DIMM_MEMDEV_PROP
,
3510 pagesize
= host_memory_backend_pagesize(MEMORY_BACKEND(memdev
));
3511 if (!spapr_check_pagesize(spapr
, pagesize
, errp
)) {
3515 pc_dimm_pre_plug(dimm
, MACHINE(hotplug_dev
), NULL
, errp
);
3518 struct SpaprDimmState
{
3521 QTAILQ_ENTRY(SpaprDimmState
) next
;
3524 static SpaprDimmState
*spapr_pending_dimm_unplugs_find(SpaprMachineState
*s
,
3527 SpaprDimmState
*dimm_state
= NULL
;
3529 QTAILQ_FOREACH(dimm_state
, &s
->pending_dimm_unplugs
, next
) {
3530 if (dimm_state
->dimm
== dimm
) {
3537 static SpaprDimmState
*spapr_pending_dimm_unplugs_add(SpaprMachineState
*spapr
,
3541 SpaprDimmState
*ds
= NULL
;
3544 * If this request is for a DIMM whose removal had failed earlier
3545 * (due to guest's refusal to remove the LMBs), we would have this
3546 * dimm already in the pending_dimm_unplugs list. In that
3547 * case don't add again.
3549 ds
= spapr_pending_dimm_unplugs_find(spapr
, dimm
);
3551 ds
= g_malloc0(sizeof(SpaprDimmState
));
3552 ds
->nr_lmbs
= nr_lmbs
;
3554 QTAILQ_INSERT_HEAD(&spapr
->pending_dimm_unplugs
, ds
, next
);
3559 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState
*spapr
,
3560 SpaprDimmState
*dimm_state
)
3562 QTAILQ_REMOVE(&spapr
->pending_dimm_unplugs
, dimm_state
, next
);
3566 static SpaprDimmState
*spapr_recover_pending_dimm_state(SpaprMachineState
*ms
,
3570 uint64_t size
= memory_device_get_region_size(MEMORY_DEVICE(dimm
),
3572 uint32_t nr_lmbs
= size
/ SPAPR_MEMORY_BLOCK_SIZE
;
3573 uint32_t avail_lmbs
= 0;
3574 uint64_t addr_start
, addr
;
3577 addr_start
= object_property_get_uint(OBJECT(dimm
), PC_DIMM_ADDR_PROP
,
3581 for (i
= 0; i
< nr_lmbs
; i
++) {
3582 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
3583 addr
/ SPAPR_MEMORY_BLOCK_SIZE
);
3588 addr
+= SPAPR_MEMORY_BLOCK_SIZE
;
3591 return spapr_pending_dimm_unplugs_add(ms
, avail_lmbs
, dimm
);
3594 void spapr_memory_unplug_rollback(SpaprMachineState
*spapr
, DeviceState
*dev
)
3600 uint64_t size
, addr_start
, addr
;
3601 g_autofree
char *qapi_error
= NULL
;
3608 dimm
= PC_DIMM(dev
);
3609 ds
= spapr_pending_dimm_unplugs_find(spapr
, dimm
);
3612 * 'ds == NULL' would mean that the DIMM doesn't have a pending
3613 * unplug state, but one of its DRC is marked as unplug_requested.
3614 * This is bad and weird enough to g_assert() out.
3618 spapr_pending_dimm_unplugs_remove(spapr
, ds
);
3620 size
= memory_device_get_region_size(MEMORY_DEVICE(dimm
), &error_abort
);
3621 nr_lmbs
= size
/ SPAPR_MEMORY_BLOCK_SIZE
;
3623 addr_start
= object_property_get_uint(OBJECT(dimm
), PC_DIMM_ADDR_PROP
,
3627 for (i
= 0; i
< nr_lmbs
; i
++) {
3628 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
3629 addr
/ SPAPR_MEMORY_BLOCK_SIZE
);
3632 drc
->unplug_requested
= false;
3633 addr
+= SPAPR_MEMORY_BLOCK_SIZE
;
3637 * Tell QAPI that something happened and the memory
3638 * hotunplug wasn't successful.
3640 qapi_error
= g_strdup_printf("Memory hotunplug rejected by the guest "
3641 "for device %s", dev
->id
);
3642 qapi_event_send_mem_unplug_error(dev
->id
, qapi_error
);
3645 /* Callback to be called during DRC release. */
3646 void spapr_lmb_release(DeviceState
*dev
)
3648 HotplugHandler
*hotplug_ctrl
= qdev_get_hotplug_handler(dev
);
3649 SpaprMachineState
*spapr
= SPAPR_MACHINE(hotplug_ctrl
);
3650 SpaprDimmState
*ds
= spapr_pending_dimm_unplugs_find(spapr
, PC_DIMM(dev
));
3652 /* This information will get lost if a migration occurs
3653 * during the unplug process. In this case recover it. */
3655 ds
= spapr_recover_pending_dimm_state(spapr
, PC_DIMM(dev
));
3657 /* The DRC being examined by the caller at least must be counted */
3658 g_assert(ds
->nr_lmbs
);
3661 if (--ds
->nr_lmbs
) {
3666 * Now that all the LMBs have been removed by the guest, call the
3667 * unplug handler chain. This can never fail.
3669 hotplug_handler_unplug(hotplug_ctrl
, dev
, &error_abort
);
3670 object_unparent(OBJECT(dev
));
3673 static void spapr_memory_unplug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
)
3675 SpaprMachineState
*spapr
= SPAPR_MACHINE(hotplug_dev
);
3676 SpaprDimmState
*ds
= spapr_pending_dimm_unplugs_find(spapr
, PC_DIMM(dev
));
3678 /* We really shouldn't get this far without anything to unplug */
3681 pc_dimm_unplug(PC_DIMM(dev
), MACHINE(hotplug_dev
));
3682 qdev_unrealize(dev
);
3683 spapr_pending_dimm_unplugs_remove(spapr
, ds
);
3686 static void spapr_memory_unplug_request(HotplugHandler
*hotplug_dev
,
3687 DeviceState
*dev
, Error
**errp
)
3689 SpaprMachineState
*spapr
= SPAPR_MACHINE(hotplug_dev
);
3690 PCDIMMDevice
*dimm
= PC_DIMM(dev
);
3692 uint64_t size
, addr_start
, addr
;
3696 if (object_dynamic_cast(OBJECT(dev
), TYPE_NVDIMM
)) {
3697 error_setg(errp
, "nvdimm device hot unplug is not supported yet.");
3701 size
= memory_device_get_region_size(MEMORY_DEVICE(dimm
), &error_abort
);
3702 nr_lmbs
= size
/ SPAPR_MEMORY_BLOCK_SIZE
;
3704 addr_start
= object_property_get_uint(OBJECT(dimm
), PC_DIMM_ADDR_PROP
,
3708 * An existing pending dimm state for this DIMM means that there is an
3709 * unplug operation in progress, waiting for the spapr_lmb_release
3710 * callback to complete the job (BQL can't cover that far). In this case,
3711 * bail out to avoid detaching DRCs that were already released.
3713 if (spapr_pending_dimm_unplugs_find(spapr
, dimm
)) {
3714 error_setg(errp
, "Memory unplug already in progress for device %s",
3719 spapr_pending_dimm_unplugs_add(spapr
, nr_lmbs
, dimm
);
3722 for (i
= 0; i
< nr_lmbs
; i
++) {
3723 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
3724 addr
/ SPAPR_MEMORY_BLOCK_SIZE
);
3727 spapr_drc_unplug_request(drc
);
3728 addr
+= SPAPR_MEMORY_BLOCK_SIZE
;
3731 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
3732 addr_start
/ SPAPR_MEMORY_BLOCK_SIZE
);
3733 spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB
,
3734 nr_lmbs
, spapr_drc_index(drc
));
3737 /* Callback to be called during DRC release. */
3738 void spapr_core_release(DeviceState
*dev
)
3740 HotplugHandler
*hotplug_ctrl
= qdev_get_hotplug_handler(dev
);
3742 /* Call the unplug handler chain. This can never fail. */
3743 hotplug_handler_unplug(hotplug_ctrl
, dev
, &error_abort
);
3744 object_unparent(OBJECT(dev
));
3747 static void spapr_core_unplug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
)
3749 MachineState
*ms
= MACHINE(hotplug_dev
);
3750 SpaprMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(ms
);
3751 CPUCore
*cc
= CPU_CORE(dev
);
3752 CPUArchId
*core_slot
= spapr_find_cpu_slot(ms
, cc
->core_id
, NULL
);
3754 if (smc
->pre_2_10_has_unused_icps
) {
3755 SpaprCpuCore
*sc
= SPAPR_CPU_CORE(OBJECT(dev
));
3758 for (i
= 0; i
< cc
->nr_threads
; i
++) {
3759 CPUState
*cs
= CPU(sc
->threads
[i
]);
3761 pre_2_10_vmstate_register_dummy_icp(cs
->cpu_index
);
3766 core_slot
->cpu
= NULL
;
3767 qdev_unrealize(dev
);
3771 void spapr_core_unplug_request(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
3774 SpaprMachineState
*spapr
= SPAPR_MACHINE(OBJECT(hotplug_dev
));
3777 CPUCore
*cc
= CPU_CORE(dev
);
3779 if (!spapr_find_cpu_slot(MACHINE(hotplug_dev
), cc
->core_id
, &index
)) {
3780 error_setg(errp
, "Unable to find CPU core with core-id: %d",
3785 error_setg(errp
, "Boot CPU core may not be unplugged");
3789 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_CPU
,
3790 spapr_vcpu_id(spapr
, cc
->core_id
));
3793 if (!spapr_drc_unplug_requested(drc
)) {
3794 spapr_drc_unplug_request(drc
);
3798 * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3799 * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3800 * pulses removing the same CPU. Otherwise, in an failed hotunplug
3801 * attempt (e.g. the kernel will refuse to remove the last online
3802 * CPU), we will never attempt it again because unplug_requested
3803 * will still be 'true' in that case.
3805 spapr_hotplug_req_remove_by_index(drc
);
3808 int spapr_core_dt_populate(SpaprDrc
*drc
, SpaprMachineState
*spapr
,
3809 void *fdt
, int *fdt_start_offset
, Error
**errp
)
3811 SpaprCpuCore
*core
= SPAPR_CPU_CORE(drc
->dev
);
3812 CPUState
*cs
= CPU(core
->threads
[0]);
3813 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
3814 DeviceClass
*dc
= DEVICE_GET_CLASS(cs
);
3815 int id
= spapr_get_vcpu_id(cpu
);
3816 g_autofree
char *nodename
= NULL
;
3819 nodename
= g_strdup_printf("%s@%x", dc
->fw_name
, id
);
3820 offset
= fdt_add_subnode(fdt
, 0, nodename
);
3822 spapr_dt_cpu(cs
, fdt
, offset
, spapr
);
3825 * spapr_dt_cpu() does not fill the 'name' property in the
3826 * CPU node. The function is called during boot process, before
3827 * and after CAS, and overwriting the 'name' property written
3828 * by SLOF is not allowed.
3830 * Write it manually after spapr_dt_cpu(). This makes the hotplug
3831 * CPUs more compatible with the coldplugged ones, which have
3832 * the 'name' property. Linux Kernel also relies on this
3833 * property to identify CPU nodes.
3835 _FDT((fdt_setprop_string(fdt
, offset
, "name", nodename
)));
3837 *fdt_start_offset
= offset
;
3841 static void spapr_core_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
)
3843 SpaprMachineState
*spapr
= SPAPR_MACHINE(OBJECT(hotplug_dev
));
3844 MachineClass
*mc
= MACHINE_GET_CLASS(spapr
);
3845 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
3846 SpaprCpuCore
*core
= SPAPR_CPU_CORE(OBJECT(dev
));
3847 CPUCore
*cc
= CPU_CORE(dev
);
3850 CPUArchId
*core_slot
;
3852 bool hotplugged
= spapr_drc_hotplugged(dev
);
3855 core_slot
= spapr_find_cpu_slot(MACHINE(hotplug_dev
), cc
->core_id
, &index
);
3856 g_assert(core_slot
); /* Already checked in spapr_core_pre_plug() */
3858 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_CPU
,
3859 spapr_vcpu_id(spapr
, cc
->core_id
));
3861 g_assert(drc
|| !mc
->has_hotpluggable_cpus
);
3865 * spapr_core_pre_plug() already buys us this is a brand new
3866 * core being plugged into a free slot. Nothing should already
3867 * be attached to the corresponding DRC.
3869 spapr_drc_attach(drc
, dev
);
3873 * Send hotplug notification interrupt to the guest only
3874 * in case of hotplugged CPUs.
3876 spapr_hotplug_req_add_by_index(drc
);
3878 spapr_drc_reset(drc
);
3882 core_slot
->cpu
= OBJECT(dev
);
3885 * Set compatibility mode to match the boot CPU, which was either set
3886 * by the machine reset code or by CAS. This really shouldn't fail at
3890 for (i
= 0; i
< cc
->nr_threads
; i
++) {
3891 ppc_set_compat(core
->threads
[i
], POWERPC_CPU(first_cpu
)->compat_pvr
,
3896 if (smc
->pre_2_10_has_unused_icps
) {
3897 for (i
= 0; i
< cc
->nr_threads
; i
++) {
3898 cs
= CPU(core
->threads
[i
]);
3899 pre_2_10_vmstate_unregister_dummy_icp(cs
->cpu_index
);
3904 static void spapr_core_pre_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
3907 MachineState
*machine
= MACHINE(OBJECT(hotplug_dev
));
3908 MachineClass
*mc
= MACHINE_GET_CLASS(hotplug_dev
);
3909 CPUCore
*cc
= CPU_CORE(dev
);
3910 const char *base_core_type
= spapr_get_cpu_core_type(machine
->cpu_type
);
3911 const char *type
= object_get_typename(OBJECT(dev
));
3912 CPUArchId
*core_slot
;
3914 unsigned int smp_threads
= machine
->smp
.threads
;
3916 if (dev
->hotplugged
&& !mc
->has_hotpluggable_cpus
) {
3917 error_setg(errp
, "CPU hotplug not supported for this machine");
3921 if (strcmp(base_core_type
, type
)) {
3922 error_setg(errp
, "CPU core type should be %s", base_core_type
);
3926 if (cc
->core_id
% smp_threads
) {
3927 error_setg(errp
, "invalid core id %d", cc
->core_id
);
3932 * In general we should have homogeneous threads-per-core, but old
3933 * (pre hotplug support) machine types allow the last core to have
3934 * reduced threads as a compatibility hack for when we allowed
3935 * total vcpus not a multiple of threads-per-core.
3937 if (mc
->has_hotpluggable_cpus
&& (cc
->nr_threads
!= smp_threads
)) {
3938 error_setg(errp
, "invalid nr-threads %d, must be %d", cc
->nr_threads
,
3943 core_slot
= spapr_find_cpu_slot(MACHINE(hotplug_dev
), cc
->core_id
, &index
);
3945 error_setg(errp
, "core id %d out of range", cc
->core_id
);
3949 if (core_slot
->cpu
) {
3950 error_setg(errp
, "core %d already populated", cc
->core_id
);
3954 numa_cpu_pre_plug(core_slot
, dev
, errp
);
3957 int spapr_phb_dt_populate(SpaprDrc
*drc
, SpaprMachineState
*spapr
,
3958 void *fdt
, int *fdt_start_offset
, Error
**errp
)
3960 SpaprPhbState
*sphb
= SPAPR_PCI_HOST_BRIDGE(drc
->dev
);
3963 intc_phandle
= spapr_irq_get_phandle(spapr
, spapr
->fdt_blob
, errp
);
3964 if (intc_phandle
<= 0) {
3968 if (spapr_dt_phb(spapr
, sphb
, intc_phandle
, fdt
, fdt_start_offset
)) {
3969 error_setg(errp
, "unable to create FDT node for PHB %d", sphb
->index
);
3973 /* generally SLOF creates these, for hotplug it's up to QEMU */
3974 _FDT(fdt_setprop_string(fdt
, *fdt_start_offset
, "name", "pci"));
3979 static bool spapr_phb_pre_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
3982 SpaprMachineState
*spapr
= SPAPR_MACHINE(OBJECT(hotplug_dev
));
3983 SpaprPhbState
*sphb
= SPAPR_PCI_HOST_BRIDGE(dev
);
3984 SpaprMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(spapr
);
3985 const unsigned windows_supported
= spapr_phb_windows_supported(sphb
);
3988 if (dev
->hotplugged
&& !smc
->dr_phb_enabled
) {
3989 error_setg(errp
, "PHB hotplug not supported for this machine");
3993 if (sphb
->index
== (uint32_t)-1) {
3994 error_setg(errp
, "\"index\" for PAPR PHB is mandatory");
3998 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_PHB
, sphb
->index
);
3999 if (drc
&& drc
->dev
) {
4000 error_setg(errp
, "PHB %d already attached", sphb
->index
);
4005 * This will check that sphb->index doesn't exceed the maximum number of
4006 * PHBs for the current machine type.
4009 smc
->phb_placement(spapr
, sphb
->index
,
4010 &sphb
->buid
, &sphb
->io_win_addr
,
4011 &sphb
->mem_win_addr
, &sphb
->mem64_win_addr
,
4012 windows_supported
, sphb
->dma_liobn
,
4013 &sphb
->nv2_gpa_win_addr
, &sphb
->nv2_atsd_win_addr
,
4017 static void spapr_phb_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
)
4019 SpaprMachineState
*spapr
= SPAPR_MACHINE(OBJECT(hotplug_dev
));
4020 SpaprMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(spapr
);
4021 SpaprPhbState
*sphb
= SPAPR_PCI_HOST_BRIDGE(dev
);
4023 bool hotplugged
= spapr_drc_hotplugged(dev
);
4025 if (!smc
->dr_phb_enabled
) {
4029 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_PHB
, sphb
->index
);
4030 /* hotplug hooks should check it's enabled before getting this far */
4033 /* spapr_phb_pre_plug() already checked the DRC is attachable */
4034 spapr_drc_attach(drc
, dev
);
4037 spapr_hotplug_req_add_by_index(drc
);
4039 spapr_drc_reset(drc
);
4043 void spapr_phb_release(DeviceState
*dev
)
4045 HotplugHandler
*hotplug_ctrl
= qdev_get_hotplug_handler(dev
);
4047 hotplug_handler_unplug(hotplug_ctrl
, dev
, &error_abort
);
4048 object_unparent(OBJECT(dev
));
4051 static void spapr_phb_unplug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
)
4053 qdev_unrealize(dev
);
4056 static void spapr_phb_unplug_request(HotplugHandler
*hotplug_dev
,
4057 DeviceState
*dev
, Error
**errp
)
4059 SpaprPhbState
*sphb
= SPAPR_PCI_HOST_BRIDGE(dev
);
4062 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_PHB
, sphb
->index
);
4065 if (!spapr_drc_unplug_requested(drc
)) {
4066 spapr_drc_unplug_request(drc
);
4067 spapr_hotplug_req_remove_by_index(drc
);
4070 "PCI Host Bridge unplug already in progress for device %s",
4076 bool spapr_tpm_proxy_pre_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
4079 SpaprMachineState
*spapr
= SPAPR_MACHINE(OBJECT(hotplug_dev
));
4081 if (spapr
->tpm_proxy
!= NULL
) {
4082 error_setg(errp
, "Only one TPM proxy can be specified for this machine");
4089 static void spapr_tpm_proxy_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
)
4091 SpaprMachineState
*spapr
= SPAPR_MACHINE(OBJECT(hotplug_dev
));
4092 SpaprTpmProxy
*tpm_proxy
= SPAPR_TPM_PROXY(dev
);
4094 /* Already checked in spapr_tpm_proxy_pre_plug() */
4095 g_assert(spapr
->tpm_proxy
== NULL
);
4097 spapr
->tpm_proxy
= tpm_proxy
;
4100 static void spapr_tpm_proxy_unplug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
)
4102 SpaprMachineState
*spapr
= SPAPR_MACHINE(OBJECT(hotplug_dev
));
4104 qdev_unrealize(dev
);
4105 object_unparent(OBJECT(dev
));
4106 spapr
->tpm_proxy
= NULL
;
4109 static void spapr_machine_device_plug(HotplugHandler
*hotplug_dev
,
4110 DeviceState
*dev
, Error
**errp
)
4112 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
)) {
4113 spapr_memory_plug(hotplug_dev
, dev
);
4114 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
4115 spapr_core_plug(hotplug_dev
, dev
);
4116 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_PCI_HOST_BRIDGE
)) {
4117 spapr_phb_plug(hotplug_dev
, dev
);
4118 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_TPM_PROXY
)) {
4119 spapr_tpm_proxy_plug(hotplug_dev
, dev
);
4123 static void spapr_machine_device_unplug(HotplugHandler
*hotplug_dev
,
4124 DeviceState
*dev
, Error
**errp
)
4126 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
)) {
4127 spapr_memory_unplug(hotplug_dev
, dev
);
4128 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
4129 spapr_core_unplug(hotplug_dev
, dev
);
4130 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_PCI_HOST_BRIDGE
)) {
4131 spapr_phb_unplug(hotplug_dev
, dev
);
4132 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_TPM_PROXY
)) {
4133 spapr_tpm_proxy_unplug(hotplug_dev
, dev
);
4137 bool spapr_memory_hot_unplug_supported(SpaprMachineState
*spapr
)
4139 return spapr_ovec_test(spapr
->ov5_cas
, OV5_HP_EVT
) ||
4141 * CAS will process all pending unplug requests.
4143 * HACK: a guest could theoretically have cleared all bits in OV5,
4144 * but none of the guests we care for do.
4146 spapr_ovec_empty(spapr
->ov5_cas
);
4149 static void spapr_machine_device_unplug_request(HotplugHandler
*hotplug_dev
,
4150 DeviceState
*dev
, Error
**errp
)
4152 SpaprMachineState
*sms
= SPAPR_MACHINE(OBJECT(hotplug_dev
));
4153 MachineClass
*mc
= MACHINE_GET_CLASS(sms
);
4154 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4156 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
)) {
4157 if (spapr_memory_hot_unplug_supported(sms
)) {
4158 spapr_memory_unplug_request(hotplug_dev
, dev
, errp
);
4160 error_setg(errp
, "Memory hot unplug not supported for this guest");
4162 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
4163 if (!mc
->has_hotpluggable_cpus
) {
4164 error_setg(errp
, "CPU hot unplug not supported on this machine");
4167 spapr_core_unplug_request(hotplug_dev
, dev
, errp
);
4168 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_PCI_HOST_BRIDGE
)) {
4169 if (!smc
->dr_phb_enabled
) {
4170 error_setg(errp
, "PHB hot unplug not supported on this machine");
4173 spapr_phb_unplug_request(hotplug_dev
, dev
, errp
);
4174 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_TPM_PROXY
)) {
4175 spapr_tpm_proxy_unplug(hotplug_dev
, dev
);
4179 static void spapr_machine_device_pre_plug(HotplugHandler
*hotplug_dev
,
4180 DeviceState
*dev
, Error
**errp
)
4182 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
)) {
4183 spapr_memory_pre_plug(hotplug_dev
, dev
, errp
);
4184 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
4185 spapr_core_pre_plug(hotplug_dev
, dev
, errp
);
4186 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_PCI_HOST_BRIDGE
)) {
4187 spapr_phb_pre_plug(hotplug_dev
, dev
, errp
);
4188 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_TPM_PROXY
)) {
4189 spapr_tpm_proxy_pre_plug(hotplug_dev
, dev
, errp
);
4193 static HotplugHandler
*spapr_get_hotplug_handler(MachineState
*machine
,
4196 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
) ||
4197 object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
) ||
4198 object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_PCI_HOST_BRIDGE
) ||
4199 object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_TPM_PROXY
)) {
4200 return HOTPLUG_HANDLER(machine
);
4202 if (object_dynamic_cast(OBJECT(dev
), TYPE_PCI_DEVICE
)) {
4203 PCIDevice
*pcidev
= PCI_DEVICE(dev
);
4204 PCIBus
*root
= pci_device_root_bus(pcidev
);
4205 SpaprPhbState
*phb
=
4206 (SpaprPhbState
*)object_dynamic_cast(OBJECT(BUS(root
)->parent
),
4207 TYPE_SPAPR_PCI_HOST_BRIDGE
);
4210 return HOTPLUG_HANDLER(phb
);
4216 static CpuInstanceProperties
4217 spapr_cpu_index_to_props(MachineState
*machine
, unsigned cpu_index
)
4219 CPUArchId
*core_slot
;
4220 MachineClass
*mc
= MACHINE_GET_CLASS(machine
);
4222 /* make sure possible_cpu are intialized */
4223 mc
->possible_cpu_arch_ids(machine
);
4224 /* get CPU core slot containing thread that matches cpu_index */
4225 core_slot
= spapr_find_cpu_slot(machine
, cpu_index
, NULL
);
4227 return core_slot
->props
;
4230 static int64_t spapr_get_default_cpu_node_id(const MachineState
*ms
, int idx
)
4232 return idx
/ ms
->smp
.cores
% ms
->numa_state
->num_nodes
;
4235 static const CPUArchIdList
*spapr_possible_cpu_arch_ids(MachineState
*machine
)
4238 unsigned int smp_threads
= machine
->smp
.threads
;
4239 unsigned int smp_cpus
= machine
->smp
.cpus
;
4240 const char *core_type
;
4241 int spapr_max_cores
= machine
->smp
.max_cpus
/ smp_threads
;
4242 MachineClass
*mc
= MACHINE_GET_CLASS(machine
);
4244 if (!mc
->has_hotpluggable_cpus
) {
4245 spapr_max_cores
= QEMU_ALIGN_UP(smp_cpus
, smp_threads
) / smp_threads
;
4247 if (machine
->possible_cpus
) {
4248 assert(machine
->possible_cpus
->len
== spapr_max_cores
);
4249 return machine
->possible_cpus
;
4252 core_type
= spapr_get_cpu_core_type(machine
->cpu_type
);
4254 error_report("Unable to find sPAPR CPU Core definition");
4258 machine
->possible_cpus
= g_malloc0(sizeof(CPUArchIdList
) +
4259 sizeof(CPUArchId
) * spapr_max_cores
);
4260 machine
->possible_cpus
->len
= spapr_max_cores
;
4261 for (i
= 0; i
< machine
->possible_cpus
->len
; i
++) {
4262 int core_id
= i
* smp_threads
;
4264 machine
->possible_cpus
->cpus
[i
].type
= core_type
;
4265 machine
->possible_cpus
->cpus
[i
].vcpus_count
= smp_threads
;
4266 machine
->possible_cpus
->cpus
[i
].arch_id
= core_id
;
4267 machine
->possible_cpus
->cpus
[i
].props
.has_core_id
= true;
4268 machine
->possible_cpus
->cpus
[i
].props
.core_id
= core_id
;
4270 return machine
->possible_cpus
;
4273 static bool spapr_phb_placement(SpaprMachineState
*spapr
, uint32_t index
,
4274 uint64_t *buid
, hwaddr
*pio
,
4275 hwaddr
*mmio32
, hwaddr
*mmio64
,
4276 unsigned n_dma
, uint32_t *liobns
,
4277 hwaddr
*nv2gpa
, hwaddr
*nv2atsd
, Error
**errp
)
4280 * New-style PHB window placement.
4282 * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4283 * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4286 * Some guest kernels can't work with MMIO windows above 1<<46
4287 * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4289 * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4290 * PHB stacked together. (32TiB+2GiB)..(32TiB+64GiB) contains the
4291 * 2GiB 32-bit MMIO windows for each PHB. Then 33..64TiB has the
4292 * 1TiB 64-bit MMIO windows for each PHB.
4294 const uint64_t base_buid
= 0x800000020000000ULL
;
4297 /* Sanity check natural alignments */
4298 QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE
% SPAPR_PCI_MEM64_WIN_SIZE
) != 0);
4299 QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT
% SPAPR_PCI_MEM64_WIN_SIZE
) != 0);
4300 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE
% SPAPR_PCI_MEM32_WIN_SIZE
) != 0);
4301 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE
% SPAPR_PCI_IO_WIN_SIZE
) != 0);
4302 /* Sanity check bounds */
4303 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS
* SPAPR_PCI_IO_WIN_SIZE
) >
4304 SPAPR_PCI_MEM32_WIN_SIZE
);
4305 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS
* SPAPR_PCI_MEM32_WIN_SIZE
) >
4306 SPAPR_PCI_MEM64_WIN_SIZE
);
4308 if (index
>= SPAPR_MAX_PHBS
) {
4309 error_setg(errp
, "\"index\" for PAPR PHB is too large (max %llu)",
4310 SPAPR_MAX_PHBS
- 1);
4314 *buid
= base_buid
+ index
;
4315 for (i
= 0; i
< n_dma
; ++i
) {
4316 liobns
[i
] = SPAPR_PCI_LIOBN(index
, i
);
4319 *pio
= SPAPR_PCI_BASE
+ index
* SPAPR_PCI_IO_WIN_SIZE
;
4320 *mmio32
= SPAPR_PCI_BASE
+ (index
+ 1) * SPAPR_PCI_MEM32_WIN_SIZE
;
4321 *mmio64
= SPAPR_PCI_BASE
+ (index
+ 1) * SPAPR_PCI_MEM64_WIN_SIZE
;
4323 *nv2gpa
= SPAPR_PCI_NV2RAM64_WIN_BASE
+ index
* SPAPR_PCI_NV2RAM64_WIN_SIZE
;
4324 *nv2atsd
= SPAPR_PCI_NV2ATSD_WIN_BASE
+ index
* SPAPR_PCI_NV2ATSD_WIN_SIZE
;
4328 static ICSState
*spapr_ics_get(XICSFabric
*dev
, int irq
)
4330 SpaprMachineState
*spapr
= SPAPR_MACHINE(dev
);
4332 return ics_valid_irq(spapr
->ics
, irq
) ? spapr
->ics
: NULL
;
4335 static void spapr_ics_resend(XICSFabric
*dev
)
4337 SpaprMachineState
*spapr
= SPAPR_MACHINE(dev
);
4339 ics_resend(spapr
->ics
);
4342 static ICPState
*spapr_icp_get(XICSFabric
*xi
, int vcpu_id
)
4344 PowerPCCPU
*cpu
= spapr_find_cpu(vcpu_id
);
4346 return cpu
? spapr_cpu_state(cpu
)->icp
: NULL
;
4349 static void spapr_pic_print_info(InterruptStatsProvider
*obj
,
4352 SpaprMachineState
*spapr
= SPAPR_MACHINE(obj
);
4354 spapr_irq_print_info(spapr
, mon
);
4355 monitor_printf(mon
, "irqchip: %s\n",
4356 kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4360 * This is a XIVE only operation
4362 static int spapr_match_nvt(XiveFabric
*xfb
, uint8_t format
,
4363 uint8_t nvt_blk
, uint32_t nvt_idx
,
4364 bool cam_ignore
, uint8_t priority
,
4365 uint32_t logic_serv
, XiveTCTXMatch
*match
)
4367 SpaprMachineState
*spapr
= SPAPR_MACHINE(xfb
);
4368 XivePresenter
*xptr
= XIVE_PRESENTER(spapr
->active_intc
);
4369 XivePresenterClass
*xpc
= XIVE_PRESENTER_GET_CLASS(xptr
);
4372 count
= xpc
->match_nvt(xptr
, format
, nvt_blk
, nvt_idx
, cam_ignore
,
4373 priority
, logic_serv
, match
);
4379 * When we implement the save and restore of the thread interrupt
4380 * contexts in the enter/exit CPU handlers of the machine and the
4381 * escalations in QEMU, we should be able to handle non dispatched
4384 * Until this is done, the sPAPR machine should find at least one
4385 * matching context always.
4388 qemu_log_mask(LOG_GUEST_ERROR
, "XIVE: NVT %x/%x is not dispatched\n",
4395 int spapr_get_vcpu_id(PowerPCCPU
*cpu
)
4397 return cpu
->vcpu_id
;
4400 bool spapr_set_vcpu_id(PowerPCCPU
*cpu
, int cpu_index
, Error
**errp
)
4402 SpaprMachineState
*spapr
= SPAPR_MACHINE(qdev_get_machine());
4403 MachineState
*ms
= MACHINE(spapr
);
4406 vcpu_id
= spapr_vcpu_id(spapr
, cpu_index
);
4408 if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id
)) {
4409 error_setg(errp
, "Can't create CPU with id %d in KVM", vcpu_id
);
4410 error_append_hint(errp
, "Adjust the number of cpus to %d "
4411 "or try to raise the number of threads per core\n",
4412 vcpu_id
* ms
->smp
.threads
/ spapr
->vsmt
);
4416 cpu
->vcpu_id
= vcpu_id
;
4420 PowerPCCPU
*spapr_find_cpu(int vcpu_id
)
4425 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
4427 if (spapr_get_vcpu_id(cpu
) == vcpu_id
) {
4435 static void spapr_cpu_exec_enter(PPCVirtualHypervisor
*vhyp
, PowerPCCPU
*cpu
)
4437 SpaprCpuState
*spapr_cpu
= spapr_cpu_state(cpu
);
4439 /* These are only called by TCG, KVM maintains dispatch state */
4441 spapr_cpu
->prod
= false;
4442 if (spapr_cpu
->vpa_addr
) {
4443 CPUState
*cs
= CPU(cpu
);
4446 dispatch
= ldl_be_phys(cs
->as
,
4447 spapr_cpu
->vpa_addr
+ VPA_DISPATCH_COUNTER
);
4449 if ((dispatch
& 1) != 0) {
4450 qemu_log_mask(LOG_GUEST_ERROR
,
4451 "VPA: incorrect dispatch counter value for "
4452 "dispatched partition %u, correcting.\n", dispatch
);
4456 spapr_cpu
->vpa_addr
+ VPA_DISPATCH_COUNTER
, dispatch
);
4460 static void spapr_cpu_exec_exit(PPCVirtualHypervisor
*vhyp
, PowerPCCPU
*cpu
)
4462 SpaprCpuState
*spapr_cpu
= spapr_cpu_state(cpu
);
4464 if (spapr_cpu
->vpa_addr
) {
4465 CPUState
*cs
= CPU(cpu
);
4468 dispatch
= ldl_be_phys(cs
->as
,
4469 spapr_cpu
->vpa_addr
+ VPA_DISPATCH_COUNTER
);
4471 if ((dispatch
& 1) != 1) {
4472 qemu_log_mask(LOG_GUEST_ERROR
,
4473 "VPA: incorrect dispatch counter value for "
4474 "preempted partition %u, correcting.\n", dispatch
);
4478 spapr_cpu
->vpa_addr
+ VPA_DISPATCH_COUNTER
, dispatch
);
4482 static void spapr_machine_class_init(ObjectClass
*oc
, void *data
)
4484 MachineClass
*mc
= MACHINE_CLASS(oc
);
4485 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(oc
);
4486 FWPathProviderClass
*fwc
= FW_PATH_PROVIDER_CLASS(oc
);
4487 NMIClass
*nc
= NMI_CLASS(oc
);
4488 HotplugHandlerClass
*hc
= HOTPLUG_HANDLER_CLASS(oc
);
4489 PPCVirtualHypervisorClass
*vhc
= PPC_VIRTUAL_HYPERVISOR_CLASS(oc
);
4490 XICSFabricClass
*xic
= XICS_FABRIC_CLASS(oc
);
4491 InterruptStatsProviderClass
*ispc
= INTERRUPT_STATS_PROVIDER_CLASS(oc
);
4492 XiveFabricClass
*xfc
= XIVE_FABRIC_CLASS(oc
);
4494 mc
->desc
= "pSeries Logical Partition (PAPR compliant)";
4495 mc
->ignore_boot_device_suffixes
= true;
4498 * We set up the default / latest behaviour here. The class_init
4499 * functions for the specific versioned machine types can override
4500 * these details for backwards compatibility
4502 mc
->init
= spapr_machine_init
;
4503 mc
->reset
= spapr_machine_reset
;
4504 mc
->block_default_type
= IF_SCSI
;
4507 * Setting max_cpus to INT32_MAX. Both KVM and TCG max_cpus values
4508 * should be limited by the host capability instead of hardcoded.
4509 * max_cpus for KVM guests will be checked in kvm_init(), and TCG
4510 * guests are welcome to have as many CPUs as the host are capable
4513 mc
->max_cpus
= INT32_MAX
;
4515 mc
->no_parallel
= 1;
4516 mc
->default_boot_order
= "";
4517 mc
->default_ram_size
= 512 * MiB
;
4518 mc
->default_ram_id
= "ppc_spapr.ram";
4519 mc
->default_display
= "std";
4520 mc
->kvm_type
= spapr_kvm_type
;
4521 machine_class_allow_dynamic_sysbus_dev(mc
, TYPE_SPAPR_PCI_HOST_BRIDGE
);
4522 mc
->pci_allow_0_address
= true;
4523 assert(!mc
->get_hotplug_handler
);
4524 mc
->get_hotplug_handler
= spapr_get_hotplug_handler
;
4525 hc
->pre_plug
= spapr_machine_device_pre_plug
;
4526 hc
->plug
= spapr_machine_device_plug
;
4527 mc
->cpu_index_to_instance_props
= spapr_cpu_index_to_props
;
4528 mc
->get_default_cpu_node_id
= spapr_get_default_cpu_node_id
;
4529 mc
->possible_cpu_arch_ids
= spapr_possible_cpu_arch_ids
;
4530 hc
->unplug_request
= spapr_machine_device_unplug_request
;
4531 hc
->unplug
= spapr_machine_device_unplug
;
4533 smc
->dr_lmb_enabled
= true;
4534 smc
->update_dt_enabled
= true;
4535 mc
->default_cpu_type
= POWERPC_CPU_TYPE_NAME("power9_v2.0");
4536 mc
->has_hotpluggable_cpus
= true;
4537 mc
->nvdimm_supported
= true;
4538 smc
->resize_hpt_default
= SPAPR_RESIZE_HPT_ENABLED
;
4539 fwc
->get_dev_path
= spapr_get_fw_dev_path
;
4540 nc
->nmi_monitor_handler
= spapr_nmi
;
4541 smc
->phb_placement
= spapr_phb_placement
;
4542 vhc
->hypercall
= emulate_spapr_hypercall
;
4543 vhc
->hpt_mask
= spapr_hpt_mask
;
4544 vhc
->map_hptes
= spapr_map_hptes
;
4545 vhc
->unmap_hptes
= spapr_unmap_hptes
;
4546 vhc
->hpte_set_c
= spapr_hpte_set_c
;
4547 vhc
->hpte_set_r
= spapr_hpte_set_r
;
4548 vhc
->get_pate
= spapr_get_pate
;
4549 vhc
->encode_hpt_for_kvm_pr
= spapr_encode_hpt_for_kvm_pr
;
4550 vhc
->cpu_exec_enter
= spapr_cpu_exec_enter
;
4551 vhc
->cpu_exec_exit
= spapr_cpu_exec_exit
;
4552 xic
->ics_get
= spapr_ics_get
;
4553 xic
->ics_resend
= spapr_ics_resend
;
4554 xic
->icp_get
= spapr_icp_get
;
4555 ispc
->print_info
= spapr_pic_print_info
;
4556 /* Force NUMA node memory size to be a multiple of
4557 * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4558 * in which LMBs are represented and hot-added
4560 mc
->numa_mem_align_shift
= 28;
4561 mc
->auto_enable_numa
= true;
4563 smc
->default_caps
.caps
[SPAPR_CAP_HTM
] = SPAPR_CAP_OFF
;
4564 smc
->default_caps
.caps
[SPAPR_CAP_VSX
] = SPAPR_CAP_ON
;
4565 smc
->default_caps
.caps
[SPAPR_CAP_DFP
] = SPAPR_CAP_ON
;
4566 smc
->default_caps
.caps
[SPAPR_CAP_CFPC
] = SPAPR_CAP_WORKAROUND
;
4567 smc
->default_caps
.caps
[SPAPR_CAP_SBBC
] = SPAPR_CAP_WORKAROUND
;
4568 smc
->default_caps
.caps
[SPAPR_CAP_IBS
] = SPAPR_CAP_WORKAROUND
;
4569 smc
->default_caps
.caps
[SPAPR_CAP_HPT_MAXPAGESIZE
] = 16; /* 64kiB */
4570 smc
->default_caps
.caps
[SPAPR_CAP_NESTED_KVM_HV
] = SPAPR_CAP_OFF
;
4571 smc
->default_caps
.caps
[SPAPR_CAP_LARGE_DECREMENTER
] = SPAPR_CAP_ON
;
4572 smc
->default_caps
.caps
[SPAPR_CAP_CCF_ASSIST
] = SPAPR_CAP_ON
;
4573 smc
->default_caps
.caps
[SPAPR_CAP_FWNMI
] = SPAPR_CAP_ON
;
4574 spapr_caps_add_properties(smc
);
4575 smc
->irq
= &spapr_irq_dual
;
4576 smc
->dr_phb_enabled
= true;
4577 smc
->linux_pci_probe
= true;
4578 smc
->smp_threads_vsmt
= true;
4579 smc
->nr_xirqs
= SPAPR_NR_XIRQS
;
4580 xfc
->match_nvt
= spapr_match_nvt
;
4583 static const TypeInfo spapr_machine_info
= {
4584 .name
= TYPE_SPAPR_MACHINE
,
4585 .parent
= TYPE_MACHINE
,
4587 .instance_size
= sizeof(SpaprMachineState
),
4588 .instance_init
= spapr_instance_init
,
4589 .instance_finalize
= spapr_machine_finalizefn
,
4590 .class_size
= sizeof(SpaprMachineClass
),
4591 .class_init
= spapr_machine_class_init
,
4592 .interfaces
= (InterfaceInfo
[]) {
4593 { TYPE_FW_PATH_PROVIDER
},
4595 { TYPE_HOTPLUG_HANDLER
},
4596 { TYPE_PPC_VIRTUAL_HYPERVISOR
},
4597 { TYPE_XICS_FABRIC
},
4598 { TYPE_INTERRUPT_STATS_PROVIDER
},
4599 { TYPE_XIVE_FABRIC
},
4604 static void spapr_machine_latest_class_options(MachineClass
*mc
)
4606 mc
->alias
= "pseries";
4607 mc
->is_default
= true;
4610 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest) \
4611 static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4614 MachineClass *mc = MACHINE_CLASS(oc); \
4615 spapr_machine_##suffix##_class_options(mc); \
4617 spapr_machine_latest_class_options(mc); \
4620 static const TypeInfo spapr_machine_##suffix##_info = { \
4621 .name = MACHINE_TYPE_NAME("pseries-" verstr), \
4622 .parent = TYPE_SPAPR_MACHINE, \
4623 .class_init = spapr_machine_##suffix##_class_init, \
4625 static void spapr_machine_register_##suffix(void) \
4627 type_register(&spapr_machine_##suffix##_info); \
4629 type_init(spapr_machine_register_##suffix)
4634 static void spapr_machine_6_1_class_options(MachineClass
*mc
)
4636 /* Defaults for the latest behaviour inherited from the base class */
4639 DEFINE_SPAPR_MACHINE(6_1
, "6.1", true);
4644 static void spapr_machine_6_0_class_options(MachineClass
*mc
)
4646 spapr_machine_6_1_class_options(mc
);
4647 compat_props_add(mc
->compat_props
, hw_compat_6_0
, hw_compat_6_0_len
);
4650 DEFINE_SPAPR_MACHINE(6_0
, "6.0", false);
4655 static void spapr_machine_5_2_class_options(MachineClass
*mc
)
4657 spapr_machine_6_0_class_options(mc
);
4658 compat_props_add(mc
->compat_props
, hw_compat_5_2
, hw_compat_5_2_len
);
4661 DEFINE_SPAPR_MACHINE(5_2
, "5.2", false);
4666 static void spapr_machine_5_1_class_options(MachineClass
*mc
)
4668 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4670 spapr_machine_5_2_class_options(mc
);
4671 compat_props_add(mc
->compat_props
, hw_compat_5_1
, hw_compat_5_1_len
);
4672 smc
->pre_5_2_numa_associativity
= true;
4675 DEFINE_SPAPR_MACHINE(5_1
, "5.1", false);
4680 static void spapr_machine_5_0_class_options(MachineClass
*mc
)
4682 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4683 static GlobalProperty compat
[] = {
4684 { TYPE_SPAPR_PCI_HOST_BRIDGE
, "pre-5.1-associativity", "on" },
4687 spapr_machine_5_1_class_options(mc
);
4688 compat_props_add(mc
->compat_props
, hw_compat_5_0
, hw_compat_5_0_len
);
4689 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
4690 mc
->numa_mem_supported
= true;
4691 smc
->pre_5_1_assoc_refpoints
= true;
4694 DEFINE_SPAPR_MACHINE(5_0
, "5.0", false);
4699 static void spapr_machine_4_2_class_options(MachineClass
*mc
)
4701 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4703 spapr_machine_5_0_class_options(mc
);
4704 compat_props_add(mc
->compat_props
, hw_compat_4_2
, hw_compat_4_2_len
);
4705 smc
->default_caps
.caps
[SPAPR_CAP_CCF_ASSIST
] = SPAPR_CAP_OFF
;
4706 smc
->default_caps
.caps
[SPAPR_CAP_FWNMI
] = SPAPR_CAP_OFF
;
4707 smc
->rma_limit
= 16 * GiB
;
4708 mc
->nvdimm_supported
= false;
4711 DEFINE_SPAPR_MACHINE(4_2
, "4.2", false);
4716 static void spapr_machine_4_1_class_options(MachineClass
*mc
)
4718 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4719 static GlobalProperty compat
[] = {
4720 /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4721 { TYPE_SPAPR_PCI_HOST_BRIDGE
, "pgsz", "0x11000" },
4724 spapr_machine_4_2_class_options(mc
);
4725 smc
->linux_pci_probe
= false;
4726 smc
->smp_threads_vsmt
= false;
4727 compat_props_add(mc
->compat_props
, hw_compat_4_1
, hw_compat_4_1_len
);
4728 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
4731 DEFINE_SPAPR_MACHINE(4_1
, "4.1", false);
4736 static bool phb_placement_4_0(SpaprMachineState
*spapr
, uint32_t index
,
4737 uint64_t *buid
, hwaddr
*pio
,
4738 hwaddr
*mmio32
, hwaddr
*mmio64
,
4739 unsigned n_dma
, uint32_t *liobns
,
4740 hwaddr
*nv2gpa
, hwaddr
*nv2atsd
, Error
**errp
)
4742 if (!spapr_phb_placement(spapr
, index
, buid
, pio
, mmio32
, mmio64
, n_dma
,
4743 liobns
, nv2gpa
, nv2atsd
, errp
)) {
4751 static void spapr_machine_4_0_class_options(MachineClass
*mc
)
4753 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4755 spapr_machine_4_1_class_options(mc
);
4756 compat_props_add(mc
->compat_props
, hw_compat_4_0
, hw_compat_4_0_len
);
4757 smc
->phb_placement
= phb_placement_4_0
;
4758 smc
->irq
= &spapr_irq_xics
;
4759 smc
->pre_4_1_migration
= true;
4762 DEFINE_SPAPR_MACHINE(4_0
, "4.0", false);
4767 static void spapr_machine_3_1_class_options(MachineClass
*mc
)
4769 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4771 spapr_machine_4_0_class_options(mc
);
4772 compat_props_add(mc
->compat_props
, hw_compat_3_1
, hw_compat_3_1_len
);
4774 mc
->default_cpu_type
= POWERPC_CPU_TYPE_NAME("power8_v2.0");
4775 smc
->update_dt_enabled
= false;
4776 smc
->dr_phb_enabled
= false;
4777 smc
->broken_host_serial_model
= true;
4778 smc
->default_caps
.caps
[SPAPR_CAP_CFPC
] = SPAPR_CAP_BROKEN
;
4779 smc
->default_caps
.caps
[SPAPR_CAP_SBBC
] = SPAPR_CAP_BROKEN
;
4780 smc
->default_caps
.caps
[SPAPR_CAP_IBS
] = SPAPR_CAP_BROKEN
;
4781 smc
->default_caps
.caps
[SPAPR_CAP_LARGE_DECREMENTER
] = SPAPR_CAP_OFF
;
4784 DEFINE_SPAPR_MACHINE(3_1
, "3.1", false);
4790 static void spapr_machine_3_0_class_options(MachineClass
*mc
)
4792 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4794 spapr_machine_3_1_class_options(mc
);
4795 compat_props_add(mc
->compat_props
, hw_compat_3_0
, hw_compat_3_0_len
);
4797 smc
->legacy_irq_allocation
= true;
4798 smc
->nr_xirqs
= 0x400;
4799 smc
->irq
= &spapr_irq_xics_legacy
;
4802 DEFINE_SPAPR_MACHINE(3_0
, "3.0", false);
4807 static void spapr_machine_2_12_class_options(MachineClass
*mc
)
4809 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4810 static GlobalProperty compat
[] = {
4811 { TYPE_POWERPC_CPU
, "pre-3.0-migration", "on" },
4812 { TYPE_SPAPR_CPU_CORE
, "pre-3.0-migration", "on" },
4815 spapr_machine_3_0_class_options(mc
);
4816 compat_props_add(mc
->compat_props
, hw_compat_2_12
, hw_compat_2_12_len
);
4817 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
4819 /* We depend on kvm_enabled() to choose a default value for the
4820 * hpt-max-page-size capability. Of course we can't do it here
4821 * because this is too early and the HW accelerator isn't initialzed
4822 * yet. Postpone this to machine init (see default_caps_with_cpu()).
4824 smc
->default_caps
.caps
[SPAPR_CAP_HPT_MAXPAGESIZE
] = 0;
4827 DEFINE_SPAPR_MACHINE(2_12
, "2.12", false);
4829 static void spapr_machine_2_12_sxxm_class_options(MachineClass
*mc
)
4831 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4833 spapr_machine_2_12_class_options(mc
);
4834 smc
->default_caps
.caps
[SPAPR_CAP_CFPC
] = SPAPR_CAP_WORKAROUND
;
4835 smc
->default_caps
.caps
[SPAPR_CAP_SBBC
] = SPAPR_CAP_WORKAROUND
;
4836 smc
->default_caps
.caps
[SPAPR_CAP_IBS
] = SPAPR_CAP_FIXED_CCD
;
4839 DEFINE_SPAPR_MACHINE(2_12_sxxm
, "2.12-sxxm", false);
4845 static void spapr_machine_2_11_class_options(MachineClass
*mc
)
4847 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4849 spapr_machine_2_12_class_options(mc
);
4850 smc
->default_caps
.caps
[SPAPR_CAP_HTM
] = SPAPR_CAP_ON
;
4851 compat_props_add(mc
->compat_props
, hw_compat_2_11
, hw_compat_2_11_len
);
4854 DEFINE_SPAPR_MACHINE(2_11
, "2.11", false);
4860 static void spapr_machine_2_10_class_options(MachineClass
*mc
)
4862 spapr_machine_2_11_class_options(mc
);
4863 compat_props_add(mc
->compat_props
, hw_compat_2_10
, hw_compat_2_10_len
);
4866 DEFINE_SPAPR_MACHINE(2_10
, "2.10", false);
4872 static void spapr_machine_2_9_class_options(MachineClass
*mc
)
4874 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4875 static GlobalProperty compat
[] = {
4876 { TYPE_POWERPC_CPU
, "pre-2.10-migration", "on" },
4879 spapr_machine_2_10_class_options(mc
);
4880 compat_props_add(mc
->compat_props
, hw_compat_2_9
, hw_compat_2_9_len
);
4881 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
4882 smc
->pre_2_10_has_unused_icps
= true;
4883 smc
->resize_hpt_default
= SPAPR_RESIZE_HPT_DISABLED
;
4886 DEFINE_SPAPR_MACHINE(2_9
, "2.9", false);
4892 static void spapr_machine_2_8_class_options(MachineClass
*mc
)
4894 static GlobalProperty compat
[] = {
4895 { TYPE_SPAPR_PCI_HOST_BRIDGE
, "pcie-extended-configuration-space", "off" },
4898 spapr_machine_2_9_class_options(mc
);
4899 compat_props_add(mc
->compat_props
, hw_compat_2_8
, hw_compat_2_8_len
);
4900 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
4901 mc
->numa_mem_align_shift
= 23;
4904 DEFINE_SPAPR_MACHINE(2_8
, "2.8", false);
4910 static bool phb_placement_2_7(SpaprMachineState
*spapr
, uint32_t index
,
4911 uint64_t *buid
, hwaddr
*pio
,
4912 hwaddr
*mmio32
, hwaddr
*mmio64
,
4913 unsigned n_dma
, uint32_t *liobns
,
4914 hwaddr
*nv2gpa
, hwaddr
*nv2atsd
, Error
**errp
)
4916 /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4917 const uint64_t base_buid
= 0x800000020000000ULL
;
4918 const hwaddr phb_spacing
= 0x1000000000ULL
; /* 64 GiB */
4919 const hwaddr mmio_offset
= 0xa0000000; /* 2 GiB + 512 MiB */
4920 const hwaddr pio_offset
= 0x80000000; /* 2 GiB */
4921 const uint32_t max_index
= 255;
4922 const hwaddr phb0_alignment
= 0x10000000000ULL
; /* 1 TiB */
4924 uint64_t ram_top
= MACHINE(spapr
)->ram_size
;
4925 hwaddr phb0_base
, phb_base
;
4928 /* Do we have device memory? */
4929 if (MACHINE(spapr
)->maxram_size
> ram_top
) {
4930 /* Can't just use maxram_size, because there may be an
4931 * alignment gap between normal and device memory regions
4933 ram_top
= MACHINE(spapr
)->device_memory
->base
+
4934 memory_region_size(&MACHINE(spapr
)->device_memory
->mr
);
4937 phb0_base
= QEMU_ALIGN_UP(ram_top
, phb0_alignment
);
4939 if (index
> max_index
) {
4940 error_setg(errp
, "\"index\" for PAPR PHB is too large (max %u)",
4945 *buid
= base_buid
+ index
;
4946 for (i
= 0; i
< n_dma
; ++i
) {
4947 liobns
[i
] = SPAPR_PCI_LIOBN(index
, i
);
4950 phb_base
= phb0_base
+ index
* phb_spacing
;
4951 *pio
= phb_base
+ pio_offset
;
4952 *mmio32
= phb_base
+ mmio_offset
;
4954 * We don't set the 64-bit MMIO window, relying on the PHB's
4955 * fallback behaviour of automatically splitting a large "32-bit"
4956 * window into contiguous 32-bit and 64-bit windows
4964 static void spapr_machine_2_7_class_options(MachineClass
*mc
)
4966 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4967 static GlobalProperty compat
[] = {
4968 { TYPE_SPAPR_PCI_HOST_BRIDGE
, "mem_win_size", "0xf80000000", },
4969 { TYPE_SPAPR_PCI_HOST_BRIDGE
, "mem64_win_size", "0", },
4970 { TYPE_POWERPC_CPU
, "pre-2.8-migration", "on", },
4971 { TYPE_SPAPR_PCI_HOST_BRIDGE
, "pre-2.8-migration", "on", },
4974 spapr_machine_2_8_class_options(mc
);
4975 mc
->default_cpu_type
= POWERPC_CPU_TYPE_NAME("power7_v2.3");
4976 mc
->default_machine_opts
= "modern-hotplug-events=off";
4977 compat_props_add(mc
->compat_props
, hw_compat_2_7
, hw_compat_2_7_len
);
4978 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
4979 smc
->phb_placement
= phb_placement_2_7
;
4982 DEFINE_SPAPR_MACHINE(2_7
, "2.7", false);
4988 static void spapr_machine_2_6_class_options(MachineClass
*mc
)
4990 static GlobalProperty compat
[] = {
4991 { TYPE_SPAPR_PCI_HOST_BRIDGE
, "ddw", "off" },
4994 spapr_machine_2_7_class_options(mc
);
4995 mc
->has_hotpluggable_cpus
= false;
4996 compat_props_add(mc
->compat_props
, hw_compat_2_6
, hw_compat_2_6_len
);
4997 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
5000 DEFINE_SPAPR_MACHINE(2_6
, "2.6", false);
5006 static void spapr_machine_2_5_class_options(MachineClass
*mc
)
5008 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
5009 static GlobalProperty compat
[] = {
5010 { "spapr-vlan", "use-rx-buffer-pools", "off" },
5013 spapr_machine_2_6_class_options(mc
);
5014 smc
->use_ohci_by_default
= true;
5015 compat_props_add(mc
->compat_props
, hw_compat_2_5
, hw_compat_2_5_len
);
5016 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
5019 DEFINE_SPAPR_MACHINE(2_5
, "2.5", false);
5025 static void spapr_machine_2_4_class_options(MachineClass
*mc
)
5027 SpaprMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
5029 spapr_machine_2_5_class_options(mc
);
5030 smc
->dr_lmb_enabled
= false;
5031 compat_props_add(mc
->compat_props
, hw_compat_2_4
, hw_compat_2_4_len
);
5034 DEFINE_SPAPR_MACHINE(2_4
, "2.4", false);
5040 static void spapr_machine_2_3_class_options(MachineClass
*mc
)
5042 static GlobalProperty compat
[] = {
5043 { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5045 spapr_machine_2_4_class_options(mc
);
5046 compat_props_add(mc
->compat_props
, hw_compat_2_3
, hw_compat_2_3_len
);
5047 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
5049 DEFINE_SPAPR_MACHINE(2_3
, "2.3", false);
5055 static void spapr_machine_2_2_class_options(MachineClass
*mc
)
5057 static GlobalProperty compat
[] = {
5058 { TYPE_SPAPR_PCI_HOST_BRIDGE
, "mem_win_size", "0x20000000" },
5061 spapr_machine_2_3_class_options(mc
);
5062 compat_props_add(mc
->compat_props
, hw_compat_2_2
, hw_compat_2_2_len
);
5063 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
5064 mc
->default_machine_opts
= "modern-hotplug-events=off,suppress-vmdesc=on";
5066 DEFINE_SPAPR_MACHINE(2_2
, "2.2", false);
5072 static void spapr_machine_2_1_class_options(MachineClass
*mc
)
5074 spapr_machine_2_2_class_options(mc
);
5075 compat_props_add(mc
->compat_props
, hw_compat_2_1
, hw_compat_2_1_len
);
5077 DEFINE_SPAPR_MACHINE(2_1
, "2.1", false);
5079 static void spapr_machine_register_types(void)
5081 type_register_static(&spapr_machine_info
);
5084 type_init(spapr_machine_register_types
)