2 * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
4 * Copyright (c) 2004-2007 Fabrice Bellard
5 * Copyright (c) 2007 Jocelyn Mayer
6 * Copyright (c) 2010 David Gibson, IBM Corporation.
8 * Permission is hereby granted, free of charge, to any person obtaining a copy
9 * of this software and associated documentation files (the "Software"), to deal
10 * in the Software without restriction, including without limitation the rights
11 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12 * copies of the Software, and to permit persons to whom the Software is
13 * furnished to do so, subject to the following conditions:
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
27 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "qapi/visitor.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/numa.h"
34 #include "hw/fw-path-provider.h"
37 #include "sysemu/device_tree.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
41 #include "migration/misc.h"
42 #include "migration/global_state.h"
43 #include "migration/register.h"
44 #include "mmu-hash64.h"
45 #include "mmu-book3s-v3.h"
46 #include "cpu-models.h"
49 #include "hw/boards.h"
50 #include "hw/ppc/ppc.h"
51 #include "hw/loader.h"
53 #include "hw/ppc/fdt.h"
54 #include "hw/ppc/spapr.h"
55 #include "hw/ppc/spapr_vio.h"
56 #include "hw/pci-host/spapr.h"
57 #include "hw/pci/msi.h"
59 #include "hw/pci/pci.h"
60 #include "hw/scsi/scsi.h"
61 #include "hw/virtio/virtio-scsi.h"
62 #include "hw/virtio/vhost-scsi-common.h"
64 #include "exec/address-spaces.h"
65 #include "exec/ram_addr.h"
67 #include "qemu/config-file.h"
68 #include "qemu/error-report.h"
71 #include "hw/intc/intc.h"
73 #include "qemu/cutils.h"
74 #include "hw/ppc/spapr_cpu_core.h"
75 #include "hw/mem/memory-device.h"
79 /* SLOF memory layout:
81 * SLOF raw image loaded at 0, copies its romfs right below the flat
82 * device-tree, then position SLOF itself 31M below that
84 * So we set FW_OVERHEAD to 40MB which should account for all of that
87 * We load our kernel at 4M, leaving space for SLOF initial image
89 #define FDT_MAX_SIZE 0x100000
90 #define RTAS_MAX_SIZE 0x10000
91 #define RTAS_MAX_ADDR 0x80000000 /* RTAS must stay below that */
92 #define FW_MAX_SIZE 0x400000
93 #define FW_FILE_NAME "slof.bin"
94 #define FW_OVERHEAD 0x2800000
95 #define KERNEL_LOAD_ADDR FW_MAX_SIZE
97 #define MIN_RMA_SLOF 128UL
99 #define PHANDLE_XICP 0x00001111
101 /* These two functions implement the VCPU id numbering: one to compute them
102 * all and one to identify thread 0 of a VCORE. Any change to the first one
103 * is likely to have an impact on the second one, so let's keep them close.
105 static int spapr_vcpu_id(sPAPRMachineState
*spapr
, int cpu_index
)
109 (cpu_index
/ smp_threads
) * spapr
->vsmt
+ cpu_index
% smp_threads
;
111 static bool spapr_is_thread0_in_vcore(sPAPRMachineState
*spapr
,
115 return spapr_get_vcpu_id(cpu
) % spapr
->vsmt
== 0;
118 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque
)
120 /* Dummy entries correspond to unused ICPState objects in older QEMUs,
121 * and newer QEMUs don't even have them. In both cases, we don't want
122 * to send anything on the wire.
127 static const VMStateDescription pre_2_10_vmstate_dummy_icp
= {
128 .name
= "icp/server",
130 .minimum_version_id
= 1,
131 .needed
= pre_2_10_vmstate_dummy_icp_needed
,
132 .fields
= (VMStateField
[]) {
133 VMSTATE_UNUSED(4), /* uint32_t xirr */
134 VMSTATE_UNUSED(1), /* uint8_t pending_priority */
135 VMSTATE_UNUSED(1), /* uint8_t mfrr */
136 VMSTATE_END_OF_LIST()
140 static void pre_2_10_vmstate_register_dummy_icp(int i
)
142 vmstate_register(NULL
, i
, &pre_2_10_vmstate_dummy_icp
,
143 (void *)(uintptr_t) i
);
146 static void pre_2_10_vmstate_unregister_dummy_icp(int i
)
148 vmstate_unregister(NULL
, &pre_2_10_vmstate_dummy_icp
,
149 (void *)(uintptr_t) i
);
152 int spapr_max_server_number(sPAPRMachineState
*spapr
)
155 return DIV_ROUND_UP(max_cpus
* spapr
->vsmt
, smp_threads
);
158 static int spapr_fixup_cpu_smt_dt(void *fdt
, int offset
, PowerPCCPU
*cpu
,
162 uint32_t servers_prop
[smt_threads
];
163 uint32_t gservers_prop
[smt_threads
* 2];
164 int index
= spapr_get_vcpu_id(cpu
);
166 if (cpu
->compat_pvr
) {
167 ret
= fdt_setprop_cell(fdt
, offset
, "cpu-version", cpu
->compat_pvr
);
173 /* Build interrupt servers and gservers properties */
174 for (i
= 0; i
< smt_threads
; i
++) {
175 servers_prop
[i
] = cpu_to_be32(index
+ i
);
176 /* Hack, direct the group queues back to cpu 0 */
177 gservers_prop
[i
*2] = cpu_to_be32(index
+ i
);
178 gservers_prop
[i
*2 + 1] = 0;
180 ret
= fdt_setprop(fdt
, offset
, "ibm,ppc-interrupt-server#s",
181 servers_prop
, sizeof(servers_prop
));
185 ret
= fdt_setprop(fdt
, offset
, "ibm,ppc-interrupt-gserver#s",
186 gservers_prop
, sizeof(gservers_prop
));
191 static int spapr_fixup_cpu_numa_dt(void *fdt
, int offset
, PowerPCCPU
*cpu
)
193 int index
= spapr_get_vcpu_id(cpu
);
194 uint32_t associativity
[] = {cpu_to_be32(0x5),
198 cpu_to_be32(cpu
->node_id
),
201 /* Advertise NUMA via ibm,associativity */
202 return fdt_setprop(fdt
, offset
, "ibm,associativity", associativity
,
203 sizeof(associativity
));
206 /* Populate the "ibm,pa-features" property */
207 static void spapr_populate_pa_features(sPAPRMachineState
*spapr
,
209 void *fdt
, int offset
,
212 uint8_t pa_features_206
[] = { 6, 0,
213 0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
214 uint8_t pa_features_207
[] = { 24, 0,
215 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
216 0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
217 0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
218 0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
219 uint8_t pa_features_300
[] = { 66, 0,
220 /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
221 /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
222 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
224 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
226 0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
227 /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
228 0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
229 /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
230 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
231 /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
232 0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
233 /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
234 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
235 /* 42: PM, 44: PC RA, 46: SC vec'd */
236 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
237 /* 48: SIMD, 50: QP BFP, 52: String */
238 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
239 /* 54: DecFP, 56: DecI, 58: SHA */
240 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
241 /* 60: NM atomic, 62: RNG */
242 0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
244 uint8_t *pa_features
= NULL
;
247 if (ppc_check_compat(cpu
, CPU_POWERPC_LOGICAL_2_06
, 0, cpu
->compat_pvr
)) {
248 pa_features
= pa_features_206
;
249 pa_size
= sizeof(pa_features_206
);
251 if (ppc_check_compat(cpu
, CPU_POWERPC_LOGICAL_2_07
, 0, cpu
->compat_pvr
)) {
252 pa_features
= pa_features_207
;
253 pa_size
= sizeof(pa_features_207
);
255 if (ppc_check_compat(cpu
, CPU_POWERPC_LOGICAL_3_00
, 0, cpu
->compat_pvr
)) {
256 pa_features
= pa_features_300
;
257 pa_size
= sizeof(pa_features_300
);
263 if (ppc_hash64_has(cpu
, PPC_HASH64_CI_LARGEPAGE
)) {
265 * Note: we keep CI large pages off by default because a 64K capable
266 * guest provisioned with large pages might otherwise try to map a qemu
267 * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
268 * even if that qemu runs on a 4k host.
269 * We dd this bit back here if we are confident this is not an issue
271 pa_features
[3] |= 0x20;
273 if ((spapr_get_cap(spapr
, SPAPR_CAP_HTM
) != 0) && pa_size
> 24) {
274 pa_features
[24] |= 0x80; /* Transactional memory support */
276 if (legacy_guest
&& pa_size
> 40) {
277 /* Workaround for broken kernels that attempt (guest) radix
278 * mode when they can't handle it, if they see the radix bit set
279 * in pa-features. So hide it from them. */
280 pa_features
[40 + 2] &= ~0x80; /* Radix MMU */
283 _FDT((fdt_setprop(fdt
, offset
, "ibm,pa-features", pa_features
, pa_size
)));
286 static int spapr_fixup_cpu_dt(void *fdt
, sPAPRMachineState
*spapr
)
288 int ret
= 0, offset
, cpus_offset
;
291 uint32_t pft_size_prop
[] = {0, cpu_to_be32(spapr
->htab_shift
)};
294 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
295 DeviceClass
*dc
= DEVICE_GET_CLASS(cs
);
296 int index
= spapr_get_vcpu_id(cpu
);
297 int compat_smt
= MIN(smp_threads
, ppc_compat_max_vthreads(cpu
));
299 if (!spapr_is_thread0_in_vcore(spapr
, cpu
)) {
303 snprintf(cpu_model
, 32, "%s@%x", dc
->fw_name
, index
);
305 cpus_offset
= fdt_path_offset(fdt
, "/cpus");
306 if (cpus_offset
< 0) {
307 cpus_offset
= fdt_add_subnode(fdt
, 0, "cpus");
308 if (cpus_offset
< 0) {
312 offset
= fdt_subnode_offset(fdt
, cpus_offset
, cpu_model
);
314 offset
= fdt_add_subnode(fdt
, cpus_offset
, cpu_model
);
320 ret
= fdt_setprop(fdt
, offset
, "ibm,pft-size",
321 pft_size_prop
, sizeof(pft_size_prop
));
326 if (nb_numa_nodes
> 1) {
327 ret
= spapr_fixup_cpu_numa_dt(fdt
, offset
, cpu
);
333 ret
= spapr_fixup_cpu_smt_dt(fdt
, offset
, cpu
, compat_smt
);
338 spapr_populate_pa_features(spapr
, cpu
, fdt
, offset
,
339 spapr
->cas_legacy_guest_workaround
);
344 static hwaddr
spapr_node0_size(MachineState
*machine
)
348 for (i
= 0; i
< nb_numa_nodes
; ++i
) {
349 if (numa_info
[i
].node_mem
) {
350 return MIN(pow2floor(numa_info
[i
].node_mem
),
355 return machine
->ram_size
;
358 static void add_str(GString
*s
, const gchar
*s1
)
360 g_string_append_len(s
, s1
, strlen(s1
) + 1);
363 static int spapr_populate_memory_node(void *fdt
, int nodeid
, hwaddr start
,
366 uint32_t associativity
[] = {
367 cpu_to_be32(0x4), /* length */
368 cpu_to_be32(0x0), cpu_to_be32(0x0),
369 cpu_to_be32(0x0), cpu_to_be32(nodeid
)
372 uint64_t mem_reg_property
[2];
375 mem_reg_property
[0] = cpu_to_be64(start
);
376 mem_reg_property
[1] = cpu_to_be64(size
);
378 sprintf(mem_name
, "memory@" TARGET_FMT_lx
, start
);
379 off
= fdt_add_subnode(fdt
, 0, mem_name
);
381 _FDT((fdt_setprop_string(fdt
, off
, "device_type", "memory")));
382 _FDT((fdt_setprop(fdt
, off
, "reg", mem_reg_property
,
383 sizeof(mem_reg_property
))));
384 _FDT((fdt_setprop(fdt
, off
, "ibm,associativity", associativity
,
385 sizeof(associativity
))));
389 static int spapr_populate_memory(sPAPRMachineState
*spapr
, void *fdt
)
391 MachineState
*machine
= MACHINE(spapr
);
392 hwaddr mem_start
, node_size
;
393 int i
, nb_nodes
= nb_numa_nodes
;
394 NodeInfo
*nodes
= numa_info
;
397 /* No NUMA nodes, assume there is just one node with whole RAM */
398 if (!nb_numa_nodes
) {
400 ramnode
.node_mem
= machine
->ram_size
;
404 for (i
= 0, mem_start
= 0; i
< nb_nodes
; ++i
) {
405 if (!nodes
[i
].node_mem
) {
408 if (mem_start
>= machine
->ram_size
) {
411 node_size
= nodes
[i
].node_mem
;
412 if (node_size
> machine
->ram_size
- mem_start
) {
413 node_size
= machine
->ram_size
- mem_start
;
417 /* spapr_machine_init() checks for rma_size <= node0_size
419 spapr_populate_memory_node(fdt
, i
, 0, spapr
->rma_size
);
420 mem_start
+= spapr
->rma_size
;
421 node_size
-= spapr
->rma_size
;
423 for ( ; node_size
; ) {
424 hwaddr sizetmp
= pow2floor(node_size
);
426 /* mem_start != 0 here */
427 if (ctzl(mem_start
) < ctzl(sizetmp
)) {
428 sizetmp
= 1ULL << ctzl(mem_start
);
431 spapr_populate_memory_node(fdt
, i
, mem_start
, sizetmp
);
432 node_size
-= sizetmp
;
433 mem_start
+= sizetmp
;
440 static void spapr_populate_cpu_dt(CPUState
*cs
, void *fdt
, int offset
,
441 sPAPRMachineState
*spapr
)
443 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
444 CPUPPCState
*env
= &cpu
->env
;
445 PowerPCCPUClass
*pcc
= POWERPC_CPU_GET_CLASS(cs
);
446 int index
= spapr_get_vcpu_id(cpu
);
447 uint32_t segs
[] = {cpu_to_be32(28), cpu_to_be32(40),
448 0xffffffff, 0xffffffff};
449 uint32_t tbfreq
= kvm_enabled() ? kvmppc_get_tbfreq()
450 : SPAPR_TIMEBASE_FREQ
;
451 uint32_t cpufreq
= kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
452 uint32_t page_sizes_prop
[64];
453 size_t page_sizes_prop_size
;
454 uint32_t vcpus_per_socket
= smp_threads
* smp_cores
;
455 uint32_t pft_size_prop
[] = {0, cpu_to_be32(spapr
->htab_shift
)};
456 int compat_smt
= MIN(smp_threads
, ppc_compat_max_vthreads(cpu
));
457 sPAPRDRConnector
*drc
;
459 uint32_t radix_AP_encodings
[PPC_PAGE_SIZES_MAX_SZ
];
462 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_CPU
, index
);
464 drc_index
= spapr_drc_index(drc
);
465 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,my-drc-index", drc_index
)));
468 _FDT((fdt_setprop_cell(fdt
, offset
, "reg", index
)));
469 _FDT((fdt_setprop_string(fdt
, offset
, "device_type", "cpu")));
471 _FDT((fdt_setprop_cell(fdt
, offset
, "cpu-version", env
->spr
[SPR_PVR
])));
472 _FDT((fdt_setprop_cell(fdt
, offset
, "d-cache-block-size",
473 env
->dcache_line_size
)));
474 _FDT((fdt_setprop_cell(fdt
, offset
, "d-cache-line-size",
475 env
->dcache_line_size
)));
476 _FDT((fdt_setprop_cell(fdt
, offset
, "i-cache-block-size",
477 env
->icache_line_size
)));
478 _FDT((fdt_setprop_cell(fdt
, offset
, "i-cache-line-size",
479 env
->icache_line_size
)));
481 if (pcc
->l1_dcache_size
) {
482 _FDT((fdt_setprop_cell(fdt
, offset
, "d-cache-size",
483 pcc
->l1_dcache_size
)));
485 warn_report("Unknown L1 dcache size for cpu");
487 if (pcc
->l1_icache_size
) {
488 _FDT((fdt_setprop_cell(fdt
, offset
, "i-cache-size",
489 pcc
->l1_icache_size
)));
491 warn_report("Unknown L1 icache size for cpu");
494 _FDT((fdt_setprop_cell(fdt
, offset
, "timebase-frequency", tbfreq
)));
495 _FDT((fdt_setprop_cell(fdt
, offset
, "clock-frequency", cpufreq
)));
496 _FDT((fdt_setprop_cell(fdt
, offset
, "slb-size", cpu
->hash64_opts
->slb_size
)));
497 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,slb-size", cpu
->hash64_opts
->slb_size
)));
498 _FDT((fdt_setprop_string(fdt
, offset
, "status", "okay")));
499 _FDT((fdt_setprop(fdt
, offset
, "64-bit", NULL
, 0)));
501 if (env
->spr_cb
[SPR_PURR
].oea_read
) {
502 _FDT((fdt_setprop(fdt
, offset
, "ibm,purr", NULL
, 0)));
505 if (ppc_hash64_has(cpu
, PPC_HASH64_1TSEG
)) {
506 _FDT((fdt_setprop(fdt
, offset
, "ibm,processor-segment-sizes",
507 segs
, sizeof(segs
))));
510 /* Advertise VSX (vector extensions) if available
511 * 1 == VMX / Altivec available
514 * Only CPUs for which we create core types in spapr_cpu_core.c
515 * are possible, and all of those have VMX */
516 if (spapr_get_cap(spapr
, SPAPR_CAP_VSX
) != 0) {
517 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,vmx", 2)));
519 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,vmx", 1)));
522 /* Advertise DFP (Decimal Floating Point) if available
523 * 0 / no property == no DFP
524 * 1 == DFP available */
525 if (spapr_get_cap(spapr
, SPAPR_CAP_DFP
) != 0) {
526 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,dfp", 1)));
529 page_sizes_prop_size
= ppc_create_page_sizes_prop(cpu
, page_sizes_prop
,
530 sizeof(page_sizes_prop
));
531 if (page_sizes_prop_size
) {
532 _FDT((fdt_setprop(fdt
, offset
, "ibm,segment-page-sizes",
533 page_sizes_prop
, page_sizes_prop_size
)));
536 spapr_populate_pa_features(spapr
, cpu
, fdt
, offset
, false);
538 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,chip-id",
539 cs
->cpu_index
/ vcpus_per_socket
)));
541 _FDT((fdt_setprop(fdt
, offset
, "ibm,pft-size",
542 pft_size_prop
, sizeof(pft_size_prop
))));
544 if (nb_numa_nodes
> 1) {
545 _FDT(spapr_fixup_cpu_numa_dt(fdt
, offset
, cpu
));
548 _FDT(spapr_fixup_cpu_smt_dt(fdt
, offset
, cpu
, compat_smt
));
550 if (pcc
->radix_page_info
) {
551 for (i
= 0; i
< pcc
->radix_page_info
->count
; i
++) {
552 radix_AP_encodings
[i
] =
553 cpu_to_be32(pcc
->radix_page_info
->entries
[i
]);
555 _FDT((fdt_setprop(fdt
, offset
, "ibm,processor-radix-AP-encodings",
557 pcc
->radix_page_info
->count
*
558 sizeof(radix_AP_encodings
[0]))));
562 static void spapr_populate_cpus_dt_node(void *fdt
, sPAPRMachineState
*spapr
)
571 cpus_offset
= fdt_add_subnode(fdt
, 0, "cpus");
573 _FDT((fdt_setprop_cell(fdt
, cpus_offset
, "#address-cells", 0x1)));
574 _FDT((fdt_setprop_cell(fdt
, cpus_offset
, "#size-cells", 0x0)));
577 * We walk the CPUs in reverse order to ensure that CPU DT nodes
578 * created by fdt_add_subnode() end up in the right order in FDT
579 * for the guest kernel the enumerate the CPUs correctly.
581 * The CPU list cannot be traversed in reverse order, so we need
587 rev
= g_renew(CPUState
*, rev
, n_cpus
+ 1);
591 for (i
= n_cpus
- 1; i
>= 0; i
--) {
592 CPUState
*cs
= rev
[i
];
593 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
594 int index
= spapr_get_vcpu_id(cpu
);
595 DeviceClass
*dc
= DEVICE_GET_CLASS(cs
);
598 if (!spapr_is_thread0_in_vcore(spapr
, cpu
)) {
602 nodename
= g_strdup_printf("%s@%x", dc
->fw_name
, index
);
603 offset
= fdt_add_subnode(fdt
, cpus_offset
, nodename
);
606 spapr_populate_cpu_dt(cs
, fdt
, offset
, spapr
);
612 static int spapr_rng_populate_dt(void *fdt
)
617 node
= qemu_fdt_add_subnode(fdt
, "/ibm,platform-facilities");
621 ret
= fdt_setprop_string(fdt
, node
, "device_type",
622 "ibm,platform-facilities");
623 ret
|= fdt_setprop_cell(fdt
, node
, "#address-cells", 0x1);
624 ret
|= fdt_setprop_cell(fdt
, node
, "#size-cells", 0x0);
626 node
= fdt_add_subnode(fdt
, node
, "ibm,random-v1");
630 ret
|= fdt_setprop_string(fdt
, node
, "compatible", "ibm,random");
635 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList
*list
, ram_addr_t addr
)
637 MemoryDeviceInfoList
*info
;
639 for (info
= list
; info
; info
= info
->next
) {
640 MemoryDeviceInfo
*value
= info
->value
;
642 if (value
&& value
->type
== MEMORY_DEVICE_INFO_KIND_DIMM
) {
643 PCDIMMDeviceInfo
*pcdimm_info
= value
->u
.dimm
.data
;
645 if (addr
>= pcdimm_info
->addr
&&
646 addr
< (pcdimm_info
->addr
+ pcdimm_info
->size
)) {
647 return pcdimm_info
->node
;
655 struct sPAPRDrconfCellV2
{
663 typedef struct DrconfCellQueue
{
664 struct sPAPRDrconfCellV2 cell
;
665 QSIMPLEQ_ENTRY(DrconfCellQueue
) entry
;
668 static DrconfCellQueue
*
669 spapr_get_drconf_cell(uint32_t seq_lmbs
, uint64_t base_addr
,
670 uint32_t drc_index
, uint32_t aa_index
,
673 DrconfCellQueue
*elem
;
675 elem
= g_malloc0(sizeof(*elem
));
676 elem
->cell
.seq_lmbs
= cpu_to_be32(seq_lmbs
);
677 elem
->cell
.base_addr
= cpu_to_be64(base_addr
);
678 elem
->cell
.drc_index
= cpu_to_be32(drc_index
);
679 elem
->cell
.aa_index
= cpu_to_be32(aa_index
);
680 elem
->cell
.flags
= cpu_to_be32(flags
);
685 /* ibm,dynamic-memory-v2 */
686 static int spapr_populate_drmem_v2(sPAPRMachineState
*spapr
, void *fdt
,
687 int offset
, MemoryDeviceInfoList
*dimms
)
689 MachineState
*machine
= MACHINE(spapr
);
690 uint8_t *int_buf
, *cur_index
, buf_len
;
692 uint64_t lmb_size
= SPAPR_MEMORY_BLOCK_SIZE
;
693 uint64_t addr
, cur_addr
, size
;
694 uint32_t nr_boot_lmbs
= (machine
->device_memory
->base
/ lmb_size
);
695 uint64_t mem_end
= machine
->device_memory
->base
+
696 memory_region_size(&machine
->device_memory
->mr
);
697 uint32_t node
, nr_entries
= 0;
698 sPAPRDRConnector
*drc
;
699 DrconfCellQueue
*elem
, *next
;
700 MemoryDeviceInfoList
*info
;
701 QSIMPLEQ_HEAD(, DrconfCellQueue
) drconf_queue
702 = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue
);
704 /* Entry to cover RAM and the gap area */
705 elem
= spapr_get_drconf_cell(nr_boot_lmbs
, 0, 0, -1,
706 SPAPR_LMB_FLAGS_RESERVED
|
707 SPAPR_LMB_FLAGS_DRC_INVALID
);
708 QSIMPLEQ_INSERT_TAIL(&drconf_queue
, elem
, entry
);
711 cur_addr
= machine
->device_memory
->base
;
712 for (info
= dimms
; info
; info
= info
->next
) {
713 PCDIMMDeviceInfo
*di
= info
->value
->u
.dimm
.data
;
719 /* Entry for hot-pluggable area */
720 if (cur_addr
< addr
) {
721 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
, cur_addr
/ lmb_size
);
723 elem
= spapr_get_drconf_cell((addr
- cur_addr
) / lmb_size
,
724 cur_addr
, spapr_drc_index(drc
), -1, 0);
725 QSIMPLEQ_INSERT_TAIL(&drconf_queue
, elem
, entry
);
730 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
, addr
/ lmb_size
);
732 elem
= spapr_get_drconf_cell(size
/ lmb_size
, addr
,
733 spapr_drc_index(drc
), node
,
734 SPAPR_LMB_FLAGS_ASSIGNED
);
735 QSIMPLEQ_INSERT_TAIL(&drconf_queue
, elem
, entry
);
737 cur_addr
= addr
+ size
;
740 /* Entry for remaining hotpluggable area */
741 if (cur_addr
< mem_end
) {
742 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
, cur_addr
/ lmb_size
);
744 elem
= spapr_get_drconf_cell((mem_end
- cur_addr
) / lmb_size
,
745 cur_addr
, spapr_drc_index(drc
), -1, 0);
746 QSIMPLEQ_INSERT_TAIL(&drconf_queue
, elem
, entry
);
750 buf_len
= nr_entries
* sizeof(struct sPAPRDrconfCellV2
) + sizeof(uint32_t);
751 int_buf
= cur_index
= g_malloc0(buf_len
);
752 *(uint32_t *)int_buf
= cpu_to_be32(nr_entries
);
753 cur_index
+= sizeof(nr_entries
);
755 QSIMPLEQ_FOREACH_SAFE(elem
, &drconf_queue
, entry
, next
) {
756 memcpy(cur_index
, &elem
->cell
, sizeof(elem
->cell
));
757 cur_index
+= sizeof(elem
->cell
);
758 QSIMPLEQ_REMOVE(&drconf_queue
, elem
, DrconfCellQueue
, entry
);
762 ret
= fdt_setprop(fdt
, offset
, "ibm,dynamic-memory-v2", int_buf
, buf_len
);
770 /* ibm,dynamic-memory */
771 static int spapr_populate_drmem_v1(sPAPRMachineState
*spapr
, void *fdt
,
772 int offset
, MemoryDeviceInfoList
*dimms
)
774 MachineState
*machine
= MACHINE(spapr
);
776 uint64_t lmb_size
= SPAPR_MEMORY_BLOCK_SIZE
;
777 uint32_t device_lmb_start
= machine
->device_memory
->base
/ lmb_size
;
778 uint32_t nr_lmbs
= (machine
->device_memory
->base
+
779 memory_region_size(&machine
->device_memory
->mr
)) /
781 uint32_t *int_buf
, *cur_index
, buf_len
;
784 * Allocate enough buffer size to fit in ibm,dynamic-memory
786 buf_len
= (nr_lmbs
* SPAPR_DR_LMB_LIST_ENTRY_SIZE
+ 1) * sizeof(uint32_t);
787 cur_index
= int_buf
= g_malloc0(buf_len
);
788 int_buf
[0] = cpu_to_be32(nr_lmbs
);
790 for (i
= 0; i
< nr_lmbs
; i
++) {
791 uint64_t addr
= i
* lmb_size
;
792 uint32_t *dynamic_memory
= cur_index
;
794 if (i
>= device_lmb_start
) {
795 sPAPRDRConnector
*drc
;
797 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
, i
);
800 dynamic_memory
[0] = cpu_to_be32(addr
>> 32);
801 dynamic_memory
[1] = cpu_to_be32(addr
& 0xffffffff);
802 dynamic_memory
[2] = cpu_to_be32(spapr_drc_index(drc
));
803 dynamic_memory
[3] = cpu_to_be32(0); /* reserved */
804 dynamic_memory
[4] = cpu_to_be32(spapr_pc_dimm_node(dimms
, addr
));
805 if (memory_region_present(get_system_memory(), addr
)) {
806 dynamic_memory
[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED
);
808 dynamic_memory
[5] = cpu_to_be32(0);
812 * LMB information for RMA, boot time RAM and gap b/n RAM and
813 * device memory region -- all these are marked as reserved
814 * and as having no valid DRC.
816 dynamic_memory
[0] = cpu_to_be32(addr
>> 32);
817 dynamic_memory
[1] = cpu_to_be32(addr
& 0xffffffff);
818 dynamic_memory
[2] = cpu_to_be32(0);
819 dynamic_memory
[3] = cpu_to_be32(0); /* reserved */
820 dynamic_memory
[4] = cpu_to_be32(-1);
821 dynamic_memory
[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED
|
822 SPAPR_LMB_FLAGS_DRC_INVALID
);
825 cur_index
+= SPAPR_DR_LMB_LIST_ENTRY_SIZE
;
827 ret
= fdt_setprop(fdt
, offset
, "ibm,dynamic-memory", int_buf
, buf_len
);
836 * Adds ibm,dynamic-reconfiguration-memory node.
837 * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
838 * of this device tree node.
840 static int spapr_populate_drconf_memory(sPAPRMachineState
*spapr
, void *fdt
)
842 MachineState
*machine
= MACHINE(spapr
);
844 uint64_t lmb_size
= SPAPR_MEMORY_BLOCK_SIZE
;
845 uint32_t prop_lmb_size
[] = {0, cpu_to_be32(lmb_size
)};
846 uint32_t *int_buf
, *cur_index
, buf_len
;
847 int nr_nodes
= nb_numa_nodes
? nb_numa_nodes
: 1;
848 MemoryDeviceInfoList
*dimms
= NULL
;
851 * Don't create the node if there is no device memory
853 if (machine
->ram_size
== machine
->maxram_size
) {
857 offset
= fdt_add_subnode(fdt
, 0, "ibm,dynamic-reconfiguration-memory");
859 ret
= fdt_setprop(fdt
, offset
, "ibm,lmb-size", prop_lmb_size
,
860 sizeof(prop_lmb_size
));
865 ret
= fdt_setprop_cell(fdt
, offset
, "ibm,memory-flags-mask", 0xff);
870 ret
= fdt_setprop_cell(fdt
, offset
, "ibm,memory-preservation-time", 0x0);
875 /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
876 dimms
= qmp_memory_device_list();
877 if (spapr_ovec_test(spapr
->ov5_cas
, OV5_DRMEM_V2
)) {
878 ret
= spapr_populate_drmem_v2(spapr
, fdt
, offset
, dimms
);
880 ret
= spapr_populate_drmem_v1(spapr
, fdt
, offset
, dimms
);
882 qapi_free_MemoryDeviceInfoList(dimms
);
888 /* ibm,associativity-lookup-arrays */
889 buf_len
= (nr_nodes
* 4 + 2) * sizeof(uint32_t);
890 cur_index
= int_buf
= g_malloc0(buf_len
);
891 int_buf
[0] = cpu_to_be32(nr_nodes
);
892 int_buf
[1] = cpu_to_be32(4); /* Number of entries per associativity list */
894 for (i
= 0; i
< nr_nodes
; i
++) {
895 uint32_t associativity
[] = {
901 memcpy(cur_index
, associativity
, sizeof(associativity
));
904 ret
= fdt_setprop(fdt
, offset
, "ibm,associativity-lookup-arrays", int_buf
,
905 (cur_index
- int_buf
) * sizeof(uint32_t));
911 static int spapr_dt_cas_updates(sPAPRMachineState
*spapr
, void *fdt
,
912 sPAPROptionVector
*ov5_updates
)
914 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(spapr
);
917 /* Generate ibm,dynamic-reconfiguration-memory node if required */
918 if (spapr_ovec_test(ov5_updates
, OV5_DRCONF_MEMORY
)) {
919 g_assert(smc
->dr_lmb_enabled
);
920 ret
= spapr_populate_drconf_memory(spapr
, fdt
);
926 offset
= fdt_path_offset(fdt
, "/chosen");
928 offset
= fdt_add_subnode(fdt
, 0, "chosen");
933 ret
= spapr_ovec_populate_dt(fdt
, offset
, spapr
->ov5_cas
,
934 "ibm,architecture-vec-5");
940 static bool spapr_hotplugged_dev_before_cas(void)
942 Object
*drc_container
, *obj
;
943 ObjectProperty
*prop
;
944 ObjectPropertyIterator iter
;
946 drc_container
= container_get(object_get_root(), "/dr-connector");
947 object_property_iter_init(&iter
, drc_container
);
948 while ((prop
= object_property_iter_next(&iter
))) {
949 if (!strstart(prop
->type
, "link<", NULL
)) {
952 obj
= object_property_get_link(drc_container
, prop
->name
, NULL
);
953 if (spapr_drc_needed(obj
)) {
960 int spapr_h_cas_compose_response(sPAPRMachineState
*spapr
,
961 target_ulong addr
, target_ulong size
,
962 sPAPROptionVector
*ov5_updates
)
964 void *fdt
, *fdt_skel
;
965 sPAPRDeviceTreeUpdateHeader hdr
= { .version_id
= 1 };
967 if (spapr_hotplugged_dev_before_cas()) {
971 if (size
< sizeof(hdr
) || size
> FW_MAX_SIZE
) {
972 error_report("SLOF provided an unexpected CAS buffer size "
973 TARGET_FMT_lu
" (min: %zu, max: %u)",
974 size
, sizeof(hdr
), FW_MAX_SIZE
);
980 /* Create skeleton */
981 fdt_skel
= g_malloc0(size
);
982 _FDT((fdt_create(fdt_skel
, size
)));
983 _FDT((fdt_finish_reservemap(fdt_skel
)));
984 _FDT((fdt_begin_node(fdt_skel
, "")));
985 _FDT((fdt_end_node(fdt_skel
)));
986 _FDT((fdt_finish(fdt_skel
)));
987 fdt
= g_malloc0(size
);
988 _FDT((fdt_open_into(fdt_skel
, fdt
, size
)));
991 /* Fixup cpu nodes */
992 _FDT((spapr_fixup_cpu_dt(fdt
, spapr
)));
994 if (spapr_dt_cas_updates(spapr
, fdt
, ov5_updates
)) {
998 /* Pack resulting tree */
999 _FDT((fdt_pack(fdt
)));
1001 if (fdt_totalsize(fdt
) + sizeof(hdr
) > size
) {
1002 trace_spapr_cas_failed(size
);
1006 cpu_physical_memory_write(addr
, &hdr
, sizeof(hdr
));
1007 cpu_physical_memory_write(addr
+ sizeof(hdr
), fdt
, fdt_totalsize(fdt
));
1008 trace_spapr_cas_continue(fdt_totalsize(fdt
) + sizeof(hdr
));
1014 static void spapr_dt_rtas(sPAPRMachineState
*spapr
, void *fdt
)
1017 GString
*hypertas
= g_string_sized_new(256);
1018 GString
*qemu_hypertas
= g_string_sized_new(256);
1019 uint32_t refpoints
[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1020 uint64_t max_device_addr
= MACHINE(spapr
)->device_memory
->base
+
1021 memory_region_size(&MACHINE(spapr
)->device_memory
->mr
);
1022 uint32_t lrdr_capacity
[] = {
1023 cpu_to_be32(max_device_addr
>> 32),
1024 cpu_to_be32(max_device_addr
& 0xffffffff),
1025 0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE
),
1026 cpu_to_be32(max_cpus
/ smp_threads
),
1028 uint32_t maxdomains
[] = {
1033 cpu_to_be32(nb_numa_nodes
? nb_numa_nodes
: 1),
1036 _FDT(rtas
= fdt_add_subnode(fdt
, 0, "rtas"));
1039 add_str(hypertas
, "hcall-pft");
1040 add_str(hypertas
, "hcall-term");
1041 add_str(hypertas
, "hcall-dabr");
1042 add_str(hypertas
, "hcall-interrupt");
1043 add_str(hypertas
, "hcall-tce");
1044 add_str(hypertas
, "hcall-vio");
1045 add_str(hypertas
, "hcall-splpar");
1046 add_str(hypertas
, "hcall-bulk");
1047 add_str(hypertas
, "hcall-set-mode");
1048 add_str(hypertas
, "hcall-sprg0");
1049 add_str(hypertas
, "hcall-copy");
1050 add_str(hypertas
, "hcall-debug");
1051 add_str(hypertas
, "hcall-vphn");
1052 add_str(qemu_hypertas
, "hcall-memop1");
1054 if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1055 add_str(hypertas
, "hcall-multi-tce");
1058 if (spapr
->resize_hpt
!= SPAPR_RESIZE_HPT_DISABLED
) {
1059 add_str(hypertas
, "hcall-hpt-resize");
1062 _FDT(fdt_setprop(fdt
, rtas
, "ibm,hypertas-functions",
1063 hypertas
->str
, hypertas
->len
));
1064 g_string_free(hypertas
, TRUE
);
1065 _FDT(fdt_setprop(fdt
, rtas
, "qemu,hypertas-functions",
1066 qemu_hypertas
->str
, qemu_hypertas
->len
));
1067 g_string_free(qemu_hypertas
, TRUE
);
1069 _FDT(fdt_setprop(fdt
, rtas
, "ibm,associativity-reference-points",
1070 refpoints
, sizeof(refpoints
)));
1072 _FDT(fdt_setprop(fdt
, rtas
, "ibm,max-associativity-domains",
1073 maxdomains
, sizeof(maxdomains
)));
1075 _FDT(fdt_setprop_cell(fdt
, rtas
, "rtas-error-log-max",
1076 RTAS_ERROR_LOG_MAX
));
1077 _FDT(fdt_setprop_cell(fdt
, rtas
, "rtas-event-scan-rate",
1078 RTAS_EVENT_SCAN_RATE
));
1080 g_assert(msi_nonbroken
);
1081 _FDT(fdt_setprop(fdt
, rtas
, "ibm,change-msix-capable", NULL
, 0));
1084 * According to PAPR, rtas ibm,os-term does not guarantee a return
1085 * back to the guest cpu.
1087 * While an additional ibm,extended-os-term property indicates
1088 * that rtas call return will always occur. Set this property.
1090 _FDT(fdt_setprop(fdt
, rtas
, "ibm,extended-os-term", NULL
, 0));
1092 _FDT(fdt_setprop(fdt
, rtas
, "ibm,lrdr-capacity",
1093 lrdr_capacity
, sizeof(lrdr_capacity
)));
1095 spapr_dt_rtas_tokens(fdt
, rtas
);
1099 * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1100 * and the XIVE features that the guest may request and thus the valid
1101 * values for bytes 23..26 of option vector 5:
1103 static void spapr_dt_ov5_platform_support(sPAPRMachineState
*spapr
, void *fdt
,
1106 PowerPCCPU
*first_ppc_cpu
= POWERPC_CPU(first_cpu
);
1109 23, spapr
->irq
->ov5
, /* Xive mode. */
1110 24, 0x00, /* Hash/Radix, filled in below. */
1111 25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1112 26, 0x40, /* Radix options: GTSE == yes. */
1115 if (!ppc_check_compat(first_ppc_cpu
, CPU_POWERPC_LOGICAL_3_00
, 0,
1116 first_ppc_cpu
->compat_pvr
)) {
1118 * If we're in a pre POWER9 compat mode then the guest should
1119 * do hash and use the legacy interrupt mode
1121 val
[1] = 0x00; /* XICS */
1122 val
[3] = 0x00; /* Hash */
1123 } else if (kvm_enabled()) {
1124 if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1125 val
[3] = 0x80; /* OV5_MMU_BOTH */
1126 } else if (kvmppc_has_cap_mmu_radix()) {
1127 val
[3] = 0x40; /* OV5_MMU_RADIX_300 */
1129 val
[3] = 0x00; /* Hash */
1132 /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1135 _FDT(fdt_setprop(fdt
, chosen
, "ibm,arch-vec-5-platform-support",
1139 static void spapr_dt_chosen(sPAPRMachineState
*spapr
, void *fdt
)
1141 MachineState
*machine
= MACHINE(spapr
);
1143 const char *boot_device
= machine
->boot_order
;
1144 char *stdout_path
= spapr_vio_stdout_path(spapr
->vio_bus
);
1146 char *bootlist
= get_boot_devices_list(&cb
);
1148 _FDT(chosen
= fdt_add_subnode(fdt
, 0, "chosen"));
1150 _FDT(fdt_setprop_string(fdt
, chosen
, "bootargs", machine
->kernel_cmdline
));
1151 _FDT(fdt_setprop_cell(fdt
, chosen
, "linux,initrd-start",
1152 spapr
->initrd_base
));
1153 _FDT(fdt_setprop_cell(fdt
, chosen
, "linux,initrd-end",
1154 spapr
->initrd_base
+ spapr
->initrd_size
));
1156 if (spapr
->kernel_size
) {
1157 uint64_t kprop
[2] = { cpu_to_be64(KERNEL_LOAD_ADDR
),
1158 cpu_to_be64(spapr
->kernel_size
) };
1160 _FDT(fdt_setprop(fdt
, chosen
, "qemu,boot-kernel",
1161 &kprop
, sizeof(kprop
)));
1162 if (spapr
->kernel_le
) {
1163 _FDT(fdt_setprop(fdt
, chosen
, "qemu,boot-kernel-le", NULL
, 0));
1167 _FDT((fdt_setprop_cell(fdt
, chosen
, "qemu,boot-menu", boot_menu
)));
1169 _FDT(fdt_setprop_cell(fdt
, chosen
, "qemu,graphic-width", graphic_width
));
1170 _FDT(fdt_setprop_cell(fdt
, chosen
, "qemu,graphic-height", graphic_height
));
1171 _FDT(fdt_setprop_cell(fdt
, chosen
, "qemu,graphic-depth", graphic_depth
));
1173 if (cb
&& bootlist
) {
1176 for (i
= 0; i
< cb
; i
++) {
1177 if (bootlist
[i
] == '\n') {
1181 _FDT(fdt_setprop_string(fdt
, chosen
, "qemu,boot-list", bootlist
));
1184 if (boot_device
&& strlen(boot_device
)) {
1185 _FDT(fdt_setprop_string(fdt
, chosen
, "qemu,boot-device", boot_device
));
1188 if (!spapr
->has_graphics
&& stdout_path
) {
1190 * "linux,stdout-path" and "stdout" properties are deprecated by linux
1191 * kernel. New platforms should only use the "stdout-path" property. Set
1192 * the new property and continue using older property to remain
1193 * compatible with the existing firmware.
1195 _FDT(fdt_setprop_string(fdt
, chosen
, "linux,stdout-path", stdout_path
));
1196 _FDT(fdt_setprop_string(fdt
, chosen
, "stdout-path", stdout_path
));
1199 spapr_dt_ov5_platform_support(spapr
, fdt
, chosen
);
1201 g_free(stdout_path
);
1205 static void spapr_dt_hypervisor(sPAPRMachineState
*spapr
, void *fdt
)
1207 /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1208 * KVM to work under pHyp with some guest co-operation */
1210 uint8_t hypercall
[16];
1212 _FDT(hypervisor
= fdt_add_subnode(fdt
, 0, "hypervisor"));
1213 /* indicate KVM hypercall interface */
1214 _FDT(fdt_setprop_string(fdt
, hypervisor
, "compatible", "linux,kvm"));
1215 if (kvmppc_has_cap_fixup_hcalls()) {
1217 * Older KVM versions with older guest kernels were broken
1218 * with the magic page, don't allow the guest to map it.
1220 if (!kvmppc_get_hypercall(first_cpu
->env_ptr
, hypercall
,
1221 sizeof(hypercall
))) {
1222 _FDT(fdt_setprop(fdt
, hypervisor
, "hcall-instructions",
1223 hypercall
, sizeof(hypercall
)));
1228 static void *spapr_build_fdt(sPAPRMachineState
*spapr
,
1232 MachineState
*machine
= MACHINE(spapr
);
1233 MachineClass
*mc
= MACHINE_GET_CLASS(machine
);
1234 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(machine
);
1240 fdt
= g_malloc0(FDT_MAX_SIZE
);
1241 _FDT((fdt_create_empty_tree(fdt
, FDT_MAX_SIZE
)));
1244 _FDT(fdt_setprop_string(fdt
, 0, "device_type", "chrp"));
1245 _FDT(fdt_setprop_string(fdt
, 0, "model", "IBM pSeries (emulated by qemu)"));
1246 _FDT(fdt_setprop_string(fdt
, 0, "compatible", "qemu,pseries"));
1249 * Add info to guest to indentify which host is it being run on
1250 * and what is the uuid of the guest
1252 if (kvmppc_get_host_model(&buf
)) {
1253 _FDT(fdt_setprop_string(fdt
, 0, "host-model", buf
));
1256 if (kvmppc_get_host_serial(&buf
)) {
1257 _FDT(fdt_setprop_string(fdt
, 0, "host-serial", buf
));
1261 buf
= qemu_uuid_unparse_strdup(&qemu_uuid
);
1263 _FDT(fdt_setprop_string(fdt
, 0, "vm,uuid", buf
));
1264 if (qemu_uuid_set
) {
1265 _FDT(fdt_setprop_string(fdt
, 0, "system-id", buf
));
1269 if (qemu_get_vm_name()) {
1270 _FDT(fdt_setprop_string(fdt
, 0, "ibm,partition-name",
1271 qemu_get_vm_name()));
1274 _FDT(fdt_setprop_cell(fdt
, 0, "#address-cells", 2));
1275 _FDT(fdt_setprop_cell(fdt
, 0, "#size-cells", 2));
1277 /* /interrupt controller */
1278 spapr
->irq
->dt_populate(spapr
, spapr_max_server_number(spapr
), fdt
,
1281 ret
= spapr_populate_memory(spapr
, fdt
);
1283 error_report("couldn't setup memory nodes in fdt");
1288 spapr_dt_vdevice(spapr
->vio_bus
, fdt
);
1290 if (object_resolve_path_type("", TYPE_SPAPR_RNG
, NULL
)) {
1291 ret
= spapr_rng_populate_dt(fdt
);
1293 error_report("could not set up rng device in the fdt");
1298 QLIST_FOREACH(phb
, &spapr
->phbs
, list
) {
1299 ret
= spapr_populate_pci_dt(phb
, PHANDLE_XICP
, fdt
,
1300 spapr
->irq
->nr_msis
);
1302 error_report("couldn't setup PCI devices in fdt");
1308 spapr_populate_cpus_dt_node(fdt
, spapr
);
1310 if (smc
->dr_lmb_enabled
) {
1311 _FDT(spapr_drc_populate_dt(fdt
, 0, NULL
, SPAPR_DR_CONNECTOR_TYPE_LMB
));
1314 if (mc
->has_hotpluggable_cpus
) {
1315 int offset
= fdt_path_offset(fdt
, "/cpus");
1316 ret
= spapr_drc_populate_dt(fdt
, offset
, NULL
,
1317 SPAPR_DR_CONNECTOR_TYPE_CPU
);
1319 error_report("Couldn't set up CPU DR device tree properties");
1324 /* /event-sources */
1325 spapr_dt_events(spapr
, fdt
);
1328 spapr_dt_rtas(spapr
, fdt
);
1331 spapr_dt_chosen(spapr
, fdt
);
1334 if (kvm_enabled()) {
1335 spapr_dt_hypervisor(spapr
, fdt
);
1338 /* Build memory reserve map */
1339 if (spapr
->kernel_size
) {
1340 _FDT((fdt_add_mem_rsv(fdt
, KERNEL_LOAD_ADDR
, spapr
->kernel_size
)));
1342 if (spapr
->initrd_size
) {
1343 _FDT((fdt_add_mem_rsv(fdt
, spapr
->initrd_base
, spapr
->initrd_size
)));
1346 /* ibm,client-architecture-support updates */
1347 ret
= spapr_dt_cas_updates(spapr
, fdt
, spapr
->ov5_cas
);
1349 error_report("couldn't setup CAS properties fdt");
1356 static uint64_t translate_kernel_address(void *opaque
, uint64_t addr
)
1358 return (addr
& 0x0fffffff) + KERNEL_LOAD_ADDR
;
1361 static void emulate_spapr_hypercall(PPCVirtualHypervisor
*vhyp
,
1364 CPUPPCState
*env
= &cpu
->env
;
1366 /* The TCG path should also be holding the BQL at this point */
1367 g_assert(qemu_mutex_iothread_locked());
1370 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1371 env
->gpr
[3] = H_PRIVILEGE
;
1373 env
->gpr
[3] = spapr_hypercall(cpu
, env
->gpr
[3], &env
->gpr
[4]);
1377 static uint64_t spapr_get_patbe(PPCVirtualHypervisor
*vhyp
)
1379 sPAPRMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1381 return spapr
->patb_entry
;
1384 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1385 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1386 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1387 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1388 #define DIRTY_HPTE(_hpte) ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1391 * Get the fd to access the kernel htab, re-opening it if necessary
1393 static int get_htab_fd(sPAPRMachineState
*spapr
)
1395 Error
*local_err
= NULL
;
1397 if (spapr
->htab_fd
>= 0) {
1398 return spapr
->htab_fd
;
1401 spapr
->htab_fd
= kvmppc_get_htab_fd(false, 0, &local_err
);
1402 if (spapr
->htab_fd
< 0) {
1403 error_report_err(local_err
);
1406 return spapr
->htab_fd
;
1409 void close_htab_fd(sPAPRMachineState
*spapr
)
1411 if (spapr
->htab_fd
>= 0) {
1412 close(spapr
->htab_fd
);
1414 spapr
->htab_fd
= -1;
1417 static hwaddr
spapr_hpt_mask(PPCVirtualHypervisor
*vhyp
)
1419 sPAPRMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1421 return HTAB_SIZE(spapr
) / HASH_PTEG_SIZE_64
- 1;
1424 static target_ulong
spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor
*vhyp
)
1426 sPAPRMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1428 assert(kvm_enabled());
1434 return (target_ulong
)(uintptr_t)spapr
->htab
| (spapr
->htab_shift
- 18);
1437 static const ppc_hash_pte64_t
*spapr_map_hptes(PPCVirtualHypervisor
*vhyp
,
1440 sPAPRMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1441 hwaddr pte_offset
= ptex
* HASH_PTE_SIZE_64
;
1445 * HTAB is controlled by KVM. Fetch into temporary buffer
1447 ppc_hash_pte64_t
*hptes
= g_malloc(n
* HASH_PTE_SIZE_64
);
1448 kvmppc_read_hptes(hptes
, ptex
, n
);
1453 * HTAB is controlled by QEMU. Just point to the internally
1456 return (const ppc_hash_pte64_t
*)(spapr
->htab
+ pte_offset
);
1459 static void spapr_unmap_hptes(PPCVirtualHypervisor
*vhyp
,
1460 const ppc_hash_pte64_t
*hptes
,
1463 sPAPRMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1466 g_free((void *)hptes
);
1469 /* Nothing to do for qemu managed HPT */
1472 static void spapr_store_hpte(PPCVirtualHypervisor
*vhyp
, hwaddr ptex
,
1473 uint64_t pte0
, uint64_t pte1
)
1475 sPAPRMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1476 hwaddr offset
= ptex
* HASH_PTE_SIZE_64
;
1479 kvmppc_write_hpte(ptex
, pte0
, pte1
);
1481 stq_p(spapr
->htab
+ offset
, pte0
);
1482 stq_p(spapr
->htab
+ offset
+ HASH_PTE_SIZE_64
/ 2, pte1
);
1486 int spapr_hpt_shift_for_ramsize(uint64_t ramsize
)
1490 /* We aim for a hash table of size 1/128 the size of RAM (rounded
1491 * up). The PAPR recommendation is actually 1/64 of RAM size, but
1492 * that's much more than is needed for Linux guests */
1493 shift
= ctz64(pow2ceil(ramsize
)) - 7;
1494 shift
= MAX(shift
, 18); /* Minimum architected size */
1495 shift
= MIN(shift
, 46); /* Maximum architected size */
1499 void spapr_free_hpt(sPAPRMachineState
*spapr
)
1501 g_free(spapr
->htab
);
1503 spapr
->htab_shift
= 0;
1504 close_htab_fd(spapr
);
1507 void spapr_reallocate_hpt(sPAPRMachineState
*spapr
, int shift
,
1512 /* Clean up any HPT info from a previous boot */
1513 spapr_free_hpt(spapr
);
1515 rc
= kvmppc_reset_htab(shift
);
1517 /* kernel-side HPT needed, but couldn't allocate one */
1518 error_setg_errno(errp
, errno
,
1519 "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1521 /* This is almost certainly fatal, but if the caller really
1522 * wants to carry on with shift == 0, it's welcome to try */
1523 } else if (rc
> 0) {
1524 /* kernel-side HPT allocated */
1527 "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1531 spapr
->htab_shift
= shift
;
1534 /* kernel-side HPT not needed, allocate in userspace instead */
1535 size_t size
= 1ULL << shift
;
1538 spapr
->htab
= qemu_memalign(size
, size
);
1540 error_setg_errno(errp
, errno
,
1541 "Could not allocate HPT of order %d", shift
);
1545 memset(spapr
->htab
, 0, size
);
1546 spapr
->htab_shift
= shift
;
1548 for (i
= 0; i
< size
/ HASH_PTE_SIZE_64
; i
++) {
1549 DIRTY_HPTE(HPTE(spapr
->htab
, i
));
1552 /* We're setting up a hash table, so that means we're not radix */
1553 spapr
->patb_entry
= 0;
1556 void spapr_setup_hpt_and_vrma(sPAPRMachineState
*spapr
)
1560 if ((spapr
->resize_hpt
== SPAPR_RESIZE_HPT_DISABLED
)
1561 || (spapr
->cas_reboot
1562 && !spapr_ovec_test(spapr
->ov5_cas
, OV5_HPT_RESIZE
))) {
1563 hpt_shift
= spapr_hpt_shift_for_ramsize(MACHINE(spapr
)->maxram_size
);
1565 uint64_t current_ram_size
;
1567 current_ram_size
= MACHINE(spapr
)->ram_size
+ get_plugged_memory_size();
1568 hpt_shift
= spapr_hpt_shift_for_ramsize(current_ram_size
);
1570 spapr_reallocate_hpt(spapr
, hpt_shift
, &error_fatal
);
1572 if (spapr
->vrma_adjust
) {
1573 spapr
->rma_size
= kvmppc_rma_size(spapr_node0_size(MACHINE(spapr
)),
1578 static int spapr_reset_drcs(Object
*child
, void *opaque
)
1580 sPAPRDRConnector
*drc
=
1581 (sPAPRDRConnector
*) object_dynamic_cast(child
,
1582 TYPE_SPAPR_DR_CONNECTOR
);
1585 spapr_drc_reset(drc
);
1591 static void spapr_machine_reset(void)
1593 MachineState
*machine
= MACHINE(qdev_get_machine());
1594 sPAPRMachineState
*spapr
= SPAPR_MACHINE(machine
);
1595 PowerPCCPU
*first_ppc_cpu
;
1596 uint32_t rtas_limit
;
1597 hwaddr rtas_addr
, fdt_addr
;
1601 spapr_caps_apply(spapr
);
1603 first_ppc_cpu
= POWERPC_CPU(first_cpu
);
1604 if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1605 ppc_type_check_compat(machine
->cpu_type
, CPU_POWERPC_LOGICAL_3_00
, 0,
1606 spapr
->max_compat_pvr
)) {
1607 /* If using KVM with radix mode available, VCPUs can be started
1608 * without a HPT because KVM will start them in radix mode.
1609 * Set the GR bit in PATB so that we know there is no HPT. */
1610 spapr
->patb_entry
= PATBE1_GR
;
1612 spapr_setup_hpt_and_vrma(spapr
);
1615 /* if this reset wasn't generated by CAS, we should reset our
1616 * negotiated options and start from scratch */
1617 if (!spapr
->cas_reboot
) {
1618 spapr_ovec_cleanup(spapr
->ov5_cas
);
1619 spapr
->ov5_cas
= spapr_ovec_new();
1621 ppc_set_compat(first_ppc_cpu
, spapr
->max_compat_pvr
, &error_fatal
);
1624 if (!SPAPR_MACHINE_GET_CLASS(spapr
)->legacy_irq_allocation
) {
1625 spapr_irq_msi_reset(spapr
);
1628 qemu_devices_reset();
1631 * This is fixing some of the default configuration of the XIVE
1632 * devices. To be called after the reset of the machine devices.
1634 spapr_irq_reset(spapr
, &error_fatal
);
1636 /* DRC reset may cause a device to be unplugged. This will cause troubles
1637 * if this device is used by another device (eg, a running vhost backend
1638 * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1639 * situations, we reset DRCs after all devices have been reset.
1641 object_child_foreach_recursive(object_get_root(), spapr_reset_drcs
, NULL
);
1643 spapr_clear_pending_events(spapr
);
1646 * We place the device tree and RTAS just below either the top of the RMA,
1647 * or just below 2GB, whichever is lowere, so that it can be
1648 * processed with 32-bit real mode code if necessary
1650 rtas_limit
= MIN(spapr
->rma_size
, RTAS_MAX_ADDR
);
1651 rtas_addr
= rtas_limit
- RTAS_MAX_SIZE
;
1652 fdt_addr
= rtas_addr
- FDT_MAX_SIZE
;
1654 fdt
= spapr_build_fdt(spapr
, rtas_addr
, spapr
->rtas_size
);
1656 spapr_load_rtas(spapr
, fdt
, rtas_addr
);
1660 /* Should only fail if we've built a corrupted tree */
1663 if (fdt_totalsize(fdt
) > FDT_MAX_SIZE
) {
1664 error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1665 fdt_totalsize(fdt
), FDT_MAX_SIZE
);
1670 qemu_fdt_dumpdtb(fdt
, fdt_totalsize(fdt
));
1671 cpu_physical_memory_write(fdt_addr
, fdt
, fdt_totalsize(fdt
));
1672 g_free(spapr
->fdt_blob
);
1673 spapr
->fdt_size
= fdt_totalsize(fdt
);
1674 spapr
->fdt_initial_size
= spapr
->fdt_size
;
1675 spapr
->fdt_blob
= fdt
;
1677 /* Set up the entry state */
1678 spapr_cpu_set_entry_state(first_ppc_cpu
, SPAPR_ENTRY_POINT
, fdt_addr
);
1679 first_ppc_cpu
->env
.gpr
[5] = 0;
1681 spapr
->cas_reboot
= false;
1684 static void spapr_create_nvram(sPAPRMachineState
*spapr
)
1686 DeviceState
*dev
= qdev_create(&spapr
->vio_bus
->bus
, "spapr-nvram");
1687 DriveInfo
*dinfo
= drive_get(IF_PFLASH
, 0, 0);
1690 qdev_prop_set_drive(dev
, "drive", blk_by_legacy_dinfo(dinfo
),
1694 qdev_init_nofail(dev
);
1696 spapr
->nvram
= (struct sPAPRNVRAM
*)dev
;
1699 static void spapr_rtc_create(sPAPRMachineState
*spapr
)
1701 object_initialize(&spapr
->rtc
, sizeof(spapr
->rtc
), TYPE_SPAPR_RTC
);
1702 object_property_add_child(OBJECT(spapr
), "rtc", OBJECT(&spapr
->rtc
),
1704 object_property_set_bool(OBJECT(&spapr
->rtc
), true, "realized",
1706 object_property_add_alias(OBJECT(spapr
), "rtc-time", OBJECT(&spapr
->rtc
),
1707 "date", &error_fatal
);
1710 /* Returns whether we want to use VGA or not */
1711 static bool spapr_vga_init(PCIBus
*pci_bus
, Error
**errp
)
1713 switch (vga_interface_type
) {
1720 return pci_vga_init(pci_bus
) != NULL
;
1723 "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1728 static int spapr_pre_load(void *opaque
)
1732 rc
= spapr_caps_pre_load(opaque
);
1740 static int spapr_post_load(void *opaque
, int version_id
)
1742 sPAPRMachineState
*spapr
= (sPAPRMachineState
*)opaque
;
1745 err
= spapr_caps_post_migration(spapr
);
1751 * In earlier versions, there was no separate qdev for the PAPR
1752 * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1753 * So when migrating from those versions, poke the incoming offset
1754 * value into the RTC device
1756 if (version_id
< 3) {
1757 err
= spapr_rtc_import_offset(&spapr
->rtc
, spapr
->rtc_offset
);
1763 if (kvm_enabled() && spapr
->patb_entry
) {
1764 PowerPCCPU
*cpu
= POWERPC_CPU(first_cpu
);
1765 bool radix
= !!(spapr
->patb_entry
& PATBE1_GR
);
1766 bool gtse
= !!(cpu
->env
.spr
[SPR_LPCR
] & LPCR_GTSE
);
1768 err
= kvmppc_configure_v3_mmu(cpu
, radix
, gtse
, spapr
->patb_entry
);
1770 error_report("Process table config unsupported by the host");
1775 err
= spapr_irq_post_load(spapr
, version_id
);
1783 static int spapr_pre_save(void *opaque
)
1787 rc
= spapr_caps_pre_save(opaque
);
1795 static bool version_before_3(void *opaque
, int version_id
)
1797 return version_id
< 3;
1800 static bool spapr_pending_events_needed(void *opaque
)
1802 sPAPRMachineState
*spapr
= (sPAPRMachineState
*)opaque
;
1803 return !QTAILQ_EMPTY(&spapr
->pending_events
);
1806 static const VMStateDescription vmstate_spapr_event_entry
= {
1807 .name
= "spapr_event_log_entry",
1809 .minimum_version_id
= 1,
1810 .fields
= (VMStateField
[]) {
1811 VMSTATE_UINT32(summary
, sPAPREventLogEntry
),
1812 VMSTATE_UINT32(extended_length
, sPAPREventLogEntry
),
1813 VMSTATE_VBUFFER_ALLOC_UINT32(extended_log
, sPAPREventLogEntry
, 0,
1814 NULL
, extended_length
),
1815 VMSTATE_END_OF_LIST()
1819 static const VMStateDescription vmstate_spapr_pending_events
= {
1820 .name
= "spapr_pending_events",
1822 .minimum_version_id
= 1,
1823 .needed
= spapr_pending_events_needed
,
1824 .fields
= (VMStateField
[]) {
1825 VMSTATE_QTAILQ_V(pending_events
, sPAPRMachineState
, 1,
1826 vmstate_spapr_event_entry
, sPAPREventLogEntry
, next
),
1827 VMSTATE_END_OF_LIST()
1831 static bool spapr_ov5_cas_needed(void *opaque
)
1833 sPAPRMachineState
*spapr
= opaque
;
1834 sPAPROptionVector
*ov5_mask
= spapr_ovec_new();
1835 sPAPROptionVector
*ov5_legacy
= spapr_ovec_new();
1836 sPAPROptionVector
*ov5_removed
= spapr_ovec_new();
1839 /* Prior to the introduction of sPAPROptionVector, we had two option
1840 * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1841 * Both of these options encode machine topology into the device-tree
1842 * in such a way that the now-booted OS should still be able to interact
1843 * appropriately with QEMU regardless of what options were actually
1844 * negotiatied on the source side.
1846 * As such, we can avoid migrating the CAS-negotiated options if these
1847 * are the only options available on the current machine/platform.
1848 * Since these are the only options available for pseries-2.7 and
1849 * earlier, this allows us to maintain old->new/new->old migration
1852 * For QEMU 2.8+, there are additional CAS-negotiatable options available
1853 * via default pseries-2.8 machines and explicit command-line parameters.
1854 * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1855 * of the actual CAS-negotiated values to continue working properly. For
1856 * example, availability of memory unplug depends on knowing whether
1857 * OV5_HP_EVT was negotiated via CAS.
1859 * Thus, for any cases where the set of available CAS-negotiatable
1860 * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1861 * include the CAS-negotiated options in the migration stream, unless
1862 * if they affect boot time behaviour only.
1864 spapr_ovec_set(ov5_mask
, OV5_FORM1_AFFINITY
);
1865 spapr_ovec_set(ov5_mask
, OV5_DRCONF_MEMORY
);
1866 spapr_ovec_set(ov5_mask
, OV5_DRMEM_V2
);
1868 /* spapr_ovec_diff returns true if bits were removed. we avoid using
1869 * the mask itself since in the future it's possible "legacy" bits may be
1870 * removed via machine options, which could generate a false positive
1871 * that breaks migration.
1873 spapr_ovec_intersect(ov5_legacy
, spapr
->ov5
, ov5_mask
);
1874 cas_needed
= spapr_ovec_diff(ov5_removed
, spapr
->ov5
, ov5_legacy
);
1876 spapr_ovec_cleanup(ov5_mask
);
1877 spapr_ovec_cleanup(ov5_legacy
);
1878 spapr_ovec_cleanup(ov5_removed
);
1883 static const VMStateDescription vmstate_spapr_ov5_cas
= {
1884 .name
= "spapr_option_vector_ov5_cas",
1886 .minimum_version_id
= 1,
1887 .needed
= spapr_ov5_cas_needed
,
1888 .fields
= (VMStateField
[]) {
1889 VMSTATE_STRUCT_POINTER_V(ov5_cas
, sPAPRMachineState
, 1,
1890 vmstate_spapr_ovec
, sPAPROptionVector
),
1891 VMSTATE_END_OF_LIST()
1895 static bool spapr_patb_entry_needed(void *opaque
)
1897 sPAPRMachineState
*spapr
= opaque
;
1899 return !!spapr
->patb_entry
;
1902 static const VMStateDescription vmstate_spapr_patb_entry
= {
1903 .name
= "spapr_patb_entry",
1905 .minimum_version_id
= 1,
1906 .needed
= spapr_patb_entry_needed
,
1907 .fields
= (VMStateField
[]) {
1908 VMSTATE_UINT64(patb_entry
, sPAPRMachineState
),
1909 VMSTATE_END_OF_LIST()
1913 static bool spapr_irq_map_needed(void *opaque
)
1915 sPAPRMachineState
*spapr
= opaque
;
1917 return spapr
->irq_map
&& !bitmap_empty(spapr
->irq_map
, spapr
->irq_map_nr
);
1920 static const VMStateDescription vmstate_spapr_irq_map
= {
1921 .name
= "spapr_irq_map",
1923 .minimum_version_id
= 1,
1924 .needed
= spapr_irq_map_needed
,
1925 .fields
= (VMStateField
[]) {
1926 VMSTATE_BITMAP(irq_map
, sPAPRMachineState
, 0, irq_map_nr
),
1927 VMSTATE_END_OF_LIST()
1931 static bool spapr_dtb_needed(void *opaque
)
1933 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(opaque
);
1935 return smc
->update_dt_enabled
;
1938 static int spapr_dtb_pre_load(void *opaque
)
1940 sPAPRMachineState
*spapr
= (sPAPRMachineState
*)opaque
;
1942 g_free(spapr
->fdt_blob
);
1943 spapr
->fdt_blob
= NULL
;
1944 spapr
->fdt_size
= 0;
1949 static const VMStateDescription vmstate_spapr_dtb
= {
1950 .name
= "spapr_dtb",
1952 .minimum_version_id
= 1,
1953 .needed
= spapr_dtb_needed
,
1954 .pre_load
= spapr_dtb_pre_load
,
1955 .fields
= (VMStateField
[]) {
1956 VMSTATE_UINT32(fdt_initial_size
, sPAPRMachineState
),
1957 VMSTATE_UINT32(fdt_size
, sPAPRMachineState
),
1958 VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob
, sPAPRMachineState
, 0, NULL
,
1960 VMSTATE_END_OF_LIST()
1964 static const VMStateDescription vmstate_spapr
= {
1967 .minimum_version_id
= 1,
1968 .pre_load
= spapr_pre_load
,
1969 .post_load
= spapr_post_load
,
1970 .pre_save
= spapr_pre_save
,
1971 .fields
= (VMStateField
[]) {
1972 /* used to be @next_irq */
1973 VMSTATE_UNUSED_BUFFER(version_before_3
, 0, 4),
1976 VMSTATE_UINT64_TEST(rtc_offset
, sPAPRMachineState
, version_before_3
),
1978 VMSTATE_PPC_TIMEBASE_V(tb
, sPAPRMachineState
, 2),
1979 VMSTATE_END_OF_LIST()
1981 .subsections
= (const VMStateDescription
*[]) {
1982 &vmstate_spapr_ov5_cas
,
1983 &vmstate_spapr_patb_entry
,
1984 &vmstate_spapr_pending_events
,
1985 &vmstate_spapr_cap_htm
,
1986 &vmstate_spapr_cap_vsx
,
1987 &vmstate_spapr_cap_dfp
,
1988 &vmstate_spapr_cap_cfpc
,
1989 &vmstate_spapr_cap_sbbc
,
1990 &vmstate_spapr_cap_ibs
,
1991 &vmstate_spapr_irq_map
,
1992 &vmstate_spapr_cap_nested_kvm_hv
,
1998 static int htab_save_setup(QEMUFile
*f
, void *opaque
)
2000 sPAPRMachineState
*spapr
= opaque
;
2002 /* "Iteration" header */
2003 if (!spapr
->htab_shift
) {
2004 qemu_put_be32(f
, -1);
2006 qemu_put_be32(f
, spapr
->htab_shift
);
2010 spapr
->htab_save_index
= 0;
2011 spapr
->htab_first_pass
= true;
2013 if (spapr
->htab_shift
) {
2014 assert(kvm_enabled());
2022 static void htab_save_chunk(QEMUFile
*f
, sPAPRMachineState
*spapr
,
2023 int chunkstart
, int n_valid
, int n_invalid
)
2025 qemu_put_be32(f
, chunkstart
);
2026 qemu_put_be16(f
, n_valid
);
2027 qemu_put_be16(f
, n_invalid
);
2028 qemu_put_buffer(f
, HPTE(spapr
->htab
, chunkstart
),
2029 HASH_PTE_SIZE_64
* n_valid
);
2032 static void htab_save_end_marker(QEMUFile
*f
)
2034 qemu_put_be32(f
, 0);
2035 qemu_put_be16(f
, 0);
2036 qemu_put_be16(f
, 0);
2039 static void htab_save_first_pass(QEMUFile
*f
, sPAPRMachineState
*spapr
,
2042 bool has_timeout
= max_ns
!= -1;
2043 int htabslots
= HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
;
2044 int index
= spapr
->htab_save_index
;
2045 int64_t starttime
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
2047 assert(spapr
->htab_first_pass
);
2052 /* Consume invalid HPTEs */
2053 while ((index
< htabslots
)
2054 && !HPTE_VALID(HPTE(spapr
->htab
, index
))) {
2055 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
2059 /* Consume valid HPTEs */
2061 while ((index
< htabslots
) && (index
- chunkstart
< USHRT_MAX
)
2062 && HPTE_VALID(HPTE(spapr
->htab
, index
))) {
2063 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
2067 if (index
> chunkstart
) {
2068 int n_valid
= index
- chunkstart
;
2070 htab_save_chunk(f
, spapr
, chunkstart
, n_valid
, 0);
2073 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - starttime
) > max_ns
) {
2077 } while ((index
< htabslots
) && !qemu_file_rate_limit(f
));
2079 if (index
>= htabslots
) {
2080 assert(index
== htabslots
);
2082 spapr
->htab_first_pass
= false;
2084 spapr
->htab_save_index
= index
;
2087 static int htab_save_later_pass(QEMUFile
*f
, sPAPRMachineState
*spapr
,
2090 bool final
= max_ns
< 0;
2091 int htabslots
= HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
;
2092 int examined
= 0, sent
= 0;
2093 int index
= spapr
->htab_save_index
;
2094 int64_t starttime
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
2096 assert(!spapr
->htab_first_pass
);
2099 int chunkstart
, invalidstart
;
2101 /* Consume non-dirty HPTEs */
2102 while ((index
< htabslots
)
2103 && !HPTE_DIRTY(HPTE(spapr
->htab
, index
))) {
2109 /* Consume valid dirty HPTEs */
2110 while ((index
< htabslots
) && (index
- chunkstart
< USHRT_MAX
)
2111 && HPTE_DIRTY(HPTE(spapr
->htab
, index
))
2112 && HPTE_VALID(HPTE(spapr
->htab
, index
))) {
2113 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
2118 invalidstart
= index
;
2119 /* Consume invalid dirty HPTEs */
2120 while ((index
< htabslots
) && (index
- invalidstart
< USHRT_MAX
)
2121 && HPTE_DIRTY(HPTE(spapr
->htab
, index
))
2122 && !HPTE_VALID(HPTE(spapr
->htab
, index
))) {
2123 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
2128 if (index
> chunkstart
) {
2129 int n_valid
= invalidstart
- chunkstart
;
2130 int n_invalid
= index
- invalidstart
;
2132 htab_save_chunk(f
, spapr
, chunkstart
, n_valid
, n_invalid
);
2133 sent
+= index
- chunkstart
;
2135 if (!final
&& (qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - starttime
) > max_ns
) {
2140 if (examined
>= htabslots
) {
2144 if (index
>= htabslots
) {
2145 assert(index
== htabslots
);
2148 } while ((examined
< htabslots
) && (!qemu_file_rate_limit(f
) || final
));
2150 if (index
>= htabslots
) {
2151 assert(index
== htabslots
);
2155 spapr
->htab_save_index
= index
;
2157 return (examined
>= htabslots
) && (sent
== 0) ? 1 : 0;
2160 #define MAX_ITERATION_NS 5000000 /* 5 ms */
2161 #define MAX_KVM_BUF_SIZE 2048
2163 static int htab_save_iterate(QEMUFile
*f
, void *opaque
)
2165 sPAPRMachineState
*spapr
= opaque
;
2169 /* Iteration header */
2170 if (!spapr
->htab_shift
) {
2171 qemu_put_be32(f
, -1);
2174 qemu_put_be32(f
, 0);
2178 assert(kvm_enabled());
2180 fd
= get_htab_fd(spapr
);
2185 rc
= kvmppc_save_htab(f
, fd
, MAX_KVM_BUF_SIZE
, MAX_ITERATION_NS
);
2189 } else if (spapr
->htab_first_pass
) {
2190 htab_save_first_pass(f
, spapr
, MAX_ITERATION_NS
);
2192 rc
= htab_save_later_pass(f
, spapr
, MAX_ITERATION_NS
);
2195 htab_save_end_marker(f
);
2200 static int htab_save_complete(QEMUFile
*f
, void *opaque
)
2202 sPAPRMachineState
*spapr
= opaque
;
2205 /* Iteration header */
2206 if (!spapr
->htab_shift
) {
2207 qemu_put_be32(f
, -1);
2210 qemu_put_be32(f
, 0);
2216 assert(kvm_enabled());
2218 fd
= get_htab_fd(spapr
);
2223 rc
= kvmppc_save_htab(f
, fd
, MAX_KVM_BUF_SIZE
, -1);
2228 if (spapr
->htab_first_pass
) {
2229 htab_save_first_pass(f
, spapr
, -1);
2231 htab_save_later_pass(f
, spapr
, -1);
2235 htab_save_end_marker(f
);
2240 static int htab_load(QEMUFile
*f
, void *opaque
, int version_id
)
2242 sPAPRMachineState
*spapr
= opaque
;
2243 uint32_t section_hdr
;
2245 Error
*local_err
= NULL
;
2247 if (version_id
< 1 || version_id
> 1) {
2248 error_report("htab_load() bad version");
2252 section_hdr
= qemu_get_be32(f
);
2254 if (section_hdr
== -1) {
2255 spapr_free_hpt(spapr
);
2260 /* First section gives the htab size */
2261 spapr_reallocate_hpt(spapr
, section_hdr
, &local_err
);
2263 error_report_err(local_err
);
2270 assert(kvm_enabled());
2272 fd
= kvmppc_get_htab_fd(true, 0, &local_err
);
2274 error_report_err(local_err
);
2281 uint16_t n_valid
, n_invalid
;
2283 index
= qemu_get_be32(f
);
2284 n_valid
= qemu_get_be16(f
);
2285 n_invalid
= qemu_get_be16(f
);
2287 if ((index
== 0) && (n_valid
== 0) && (n_invalid
== 0)) {
2292 if ((index
+ n_valid
+ n_invalid
) >
2293 (HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
)) {
2294 /* Bad index in stream */
2296 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2297 index
, n_valid
, n_invalid
, spapr
->htab_shift
);
2303 qemu_get_buffer(f
, HPTE(spapr
->htab
, index
),
2304 HASH_PTE_SIZE_64
* n_valid
);
2307 memset(HPTE(spapr
->htab
, index
+ n_valid
), 0,
2308 HASH_PTE_SIZE_64
* n_invalid
);
2315 rc
= kvmppc_load_htab_chunk(f
, fd
, index
, n_valid
, n_invalid
);
2330 static void htab_save_cleanup(void *opaque
)
2332 sPAPRMachineState
*spapr
= opaque
;
2334 close_htab_fd(spapr
);
2337 static SaveVMHandlers savevm_htab_handlers
= {
2338 .save_setup
= htab_save_setup
,
2339 .save_live_iterate
= htab_save_iterate
,
2340 .save_live_complete_precopy
= htab_save_complete
,
2341 .save_cleanup
= htab_save_cleanup
,
2342 .load_state
= htab_load
,
2345 static void spapr_boot_set(void *opaque
, const char *boot_device
,
2348 MachineState
*machine
= MACHINE(opaque
);
2349 machine
->boot_order
= g_strdup(boot_device
);
2352 static void spapr_create_lmb_dr_connectors(sPAPRMachineState
*spapr
)
2354 MachineState
*machine
= MACHINE(spapr
);
2355 uint64_t lmb_size
= SPAPR_MEMORY_BLOCK_SIZE
;
2356 uint32_t nr_lmbs
= (machine
->maxram_size
- machine
->ram_size
)/lmb_size
;
2359 for (i
= 0; i
< nr_lmbs
; i
++) {
2362 addr
= i
* lmb_size
+ machine
->device_memory
->base
;
2363 spapr_dr_connector_new(OBJECT(spapr
), TYPE_SPAPR_DRC_LMB
,
2369 * If RAM size, maxmem size and individual node mem sizes aren't aligned
2370 * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2371 * since we can't support such unaligned sizes with DRCONF_MEMORY.
2373 static void spapr_validate_node_memory(MachineState
*machine
, Error
**errp
)
2377 if (machine
->ram_size
% SPAPR_MEMORY_BLOCK_SIZE
) {
2378 error_setg(errp
, "Memory size 0x" RAM_ADDR_FMT
2379 " is not aligned to %" PRIu64
" MiB",
2381 SPAPR_MEMORY_BLOCK_SIZE
/ MiB
);
2385 if (machine
->maxram_size
% SPAPR_MEMORY_BLOCK_SIZE
) {
2386 error_setg(errp
, "Maximum memory size 0x" RAM_ADDR_FMT
2387 " is not aligned to %" PRIu64
" MiB",
2389 SPAPR_MEMORY_BLOCK_SIZE
/ MiB
);
2393 for (i
= 0; i
< nb_numa_nodes
; i
++) {
2394 if (numa_info
[i
].node_mem
% SPAPR_MEMORY_BLOCK_SIZE
) {
2396 "Node %d memory size 0x%" PRIx64
2397 " is not aligned to %" PRIu64
" MiB",
2398 i
, numa_info
[i
].node_mem
,
2399 SPAPR_MEMORY_BLOCK_SIZE
/ MiB
);
2405 /* find cpu slot in machine->possible_cpus by core_id */
2406 static CPUArchId
*spapr_find_cpu_slot(MachineState
*ms
, uint32_t id
, int *idx
)
2408 int index
= id
/ smp_threads
;
2410 if (index
>= ms
->possible_cpus
->len
) {
2416 return &ms
->possible_cpus
->cpus
[index
];
2419 static void spapr_set_vsmt_mode(sPAPRMachineState
*spapr
, Error
**errp
)
2421 Error
*local_err
= NULL
;
2422 bool vsmt_user
= !!spapr
->vsmt
;
2423 int kvm_smt
= kvmppc_smt_threads();
2426 if (!kvm_enabled() && (smp_threads
> 1)) {
2427 error_setg(&local_err
, "TCG cannot support more than 1 thread/core "
2428 "on a pseries machine");
2431 if (!is_power_of_2(smp_threads
)) {
2432 error_setg(&local_err
, "Cannot support %d threads/core on a pseries "
2433 "machine because it must be a power of 2", smp_threads
);
2437 /* Detemine the VSMT mode to use: */
2439 if (spapr
->vsmt
< smp_threads
) {
2440 error_setg(&local_err
, "Cannot support VSMT mode %d"
2441 " because it must be >= threads/core (%d)",
2442 spapr
->vsmt
, smp_threads
);
2445 /* In this case, spapr->vsmt has been set by the command line */
2448 * Default VSMT value is tricky, because we need it to be as
2449 * consistent as possible (for migration), but this requires
2450 * changing it for at least some existing cases. We pick 8 as
2451 * the value that we'd get with KVM on POWER8, the
2452 * overwhelmingly common case in production systems.
2454 spapr
->vsmt
= MAX(8, smp_threads
);
2457 /* KVM: If necessary, set the SMT mode: */
2458 if (kvm_enabled() && (spapr
->vsmt
!= kvm_smt
)) {
2459 ret
= kvmppc_set_smt_threads(spapr
->vsmt
);
2461 /* Looks like KVM isn't able to change VSMT mode */
2462 error_setg(&local_err
,
2463 "Failed to set KVM's VSMT mode to %d (errno %d)",
2465 /* We can live with that if the default one is big enough
2466 * for the number of threads, and a submultiple of the one
2467 * we want. In this case we'll waste some vcpu ids, but
2468 * behaviour will be correct */
2469 if ((kvm_smt
>= smp_threads
) && ((spapr
->vsmt
% kvm_smt
) == 0)) {
2470 warn_report_err(local_err
);
2475 error_append_hint(&local_err
,
2476 "On PPC, a VM with %d threads/core"
2477 " on a host with %d threads/core"
2478 " requires the use of VSMT mode %d.\n",
2479 smp_threads
, kvm_smt
, spapr
->vsmt
);
2481 kvmppc_hint_smt_possible(&local_err
);
2486 /* else TCG: nothing to do currently */
2488 error_propagate(errp
, local_err
);
2491 static void spapr_init_cpus(sPAPRMachineState
*spapr
)
2493 MachineState
*machine
= MACHINE(spapr
);
2494 MachineClass
*mc
= MACHINE_GET_CLASS(machine
);
2495 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(machine
);
2496 const char *type
= spapr_get_cpu_core_type(machine
->cpu_type
);
2497 const CPUArchIdList
*possible_cpus
;
2498 int boot_cores_nr
= smp_cpus
/ smp_threads
;
2501 possible_cpus
= mc
->possible_cpu_arch_ids(machine
);
2502 if (mc
->has_hotpluggable_cpus
) {
2503 if (smp_cpus
% smp_threads
) {
2504 error_report("smp_cpus (%u) must be multiple of threads (%u)",
2505 smp_cpus
, smp_threads
);
2508 if (max_cpus
% smp_threads
) {
2509 error_report("max_cpus (%u) must be multiple of threads (%u)",
2510 max_cpus
, smp_threads
);
2514 if (max_cpus
!= smp_cpus
) {
2515 error_report("This machine version does not support CPU hotplug");
2518 boot_cores_nr
= possible_cpus
->len
;
2521 if (smc
->pre_2_10_has_unused_icps
) {
2524 for (i
= 0; i
< spapr_max_server_number(spapr
); i
++) {
2525 /* Dummy entries get deregistered when real ICPState objects
2526 * are registered during CPU core hotplug.
2528 pre_2_10_vmstate_register_dummy_icp(i
);
2532 for (i
= 0; i
< possible_cpus
->len
; i
++) {
2533 int core_id
= i
* smp_threads
;
2535 if (mc
->has_hotpluggable_cpus
) {
2536 spapr_dr_connector_new(OBJECT(spapr
), TYPE_SPAPR_DRC_CPU
,
2537 spapr_vcpu_id(spapr
, core_id
));
2540 if (i
< boot_cores_nr
) {
2541 Object
*core
= object_new(type
);
2542 int nr_threads
= smp_threads
;
2544 /* Handle the partially filled core for older machine types */
2545 if ((i
+ 1) * smp_threads
>= smp_cpus
) {
2546 nr_threads
= smp_cpus
- i
* smp_threads
;
2549 object_property_set_int(core
, nr_threads
, "nr-threads",
2551 object_property_set_int(core
, core_id
, CPU_CORE_PROP_CORE_ID
,
2553 object_property_set_bool(core
, true, "realized", &error_fatal
);
2560 static PCIHostState
*spapr_create_default_phb(void)
2564 dev
= qdev_create(NULL
, TYPE_SPAPR_PCI_HOST_BRIDGE
);
2565 qdev_prop_set_uint32(dev
, "index", 0);
2566 qdev_init_nofail(dev
);
2568 return PCI_HOST_BRIDGE(dev
);
2571 /* pSeries LPAR / sPAPR hardware init */
2572 static void spapr_machine_init(MachineState
*machine
)
2574 sPAPRMachineState
*spapr
= SPAPR_MACHINE(machine
);
2575 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(machine
);
2576 const char *kernel_filename
= machine
->kernel_filename
;
2577 const char *initrd_filename
= machine
->initrd_filename
;
2580 MemoryRegion
*sysmem
= get_system_memory();
2581 MemoryRegion
*ram
= g_new(MemoryRegion
, 1);
2582 hwaddr node0_size
= spapr_node0_size(machine
);
2583 long load_limit
, fw_size
;
2585 Error
*resize_hpt_err
= NULL
;
2587 msi_nonbroken
= true;
2589 QLIST_INIT(&spapr
->phbs
);
2590 QTAILQ_INIT(&spapr
->pending_dimm_unplugs
);
2592 /* Determine capabilities to run with */
2593 spapr_caps_init(spapr
);
2595 kvmppc_check_papr_resize_hpt(&resize_hpt_err
);
2596 if (spapr
->resize_hpt
== SPAPR_RESIZE_HPT_DEFAULT
) {
2598 * If the user explicitly requested a mode we should either
2599 * supply it, or fail completely (which we do below). But if
2600 * it's not set explicitly, we reset our mode to something
2603 if (resize_hpt_err
) {
2604 spapr
->resize_hpt
= SPAPR_RESIZE_HPT_DISABLED
;
2605 error_free(resize_hpt_err
);
2606 resize_hpt_err
= NULL
;
2608 spapr
->resize_hpt
= smc
->resize_hpt_default
;
2612 assert(spapr
->resize_hpt
!= SPAPR_RESIZE_HPT_DEFAULT
);
2614 if ((spapr
->resize_hpt
!= SPAPR_RESIZE_HPT_DISABLED
) && resize_hpt_err
) {
2616 * User requested HPT resize, but this host can't supply it. Bail out
2618 error_report_err(resize_hpt_err
);
2622 spapr
->rma_size
= node0_size
;
2624 /* With KVM, we don't actually know whether KVM supports an
2625 * unbounded RMA (PR KVM) or is limited by the hash table size
2626 * (HV KVM using VRMA), so we always assume the latter
2628 * In that case, we also limit the initial allocations for RTAS
2629 * etc... to 256M since we have no way to know what the VRMA size
2630 * is going to be as it depends on the size of the hash table
2631 * which isn't determined yet.
2633 if (kvm_enabled()) {
2634 spapr
->vrma_adjust
= 1;
2635 spapr
->rma_size
= MIN(spapr
->rma_size
, 0x10000000);
2638 /* Actually we don't support unbounded RMA anymore since we added
2639 * proper emulation of HV mode. The max we can get is 16G which
2640 * also happens to be what we configure for PAPR mode so make sure
2641 * we don't do anything bigger than that
2643 spapr
->rma_size
= MIN(spapr
->rma_size
, 0x400000000ull
);
2645 if (spapr
->rma_size
> node0_size
) {
2646 error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx
")",
2651 /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2652 load_limit
= MIN(spapr
->rma_size
, RTAS_MAX_ADDR
) - FW_OVERHEAD
;
2655 * VSMT must be set in order to be able to compute VCPU ids, ie to
2656 * call spapr_max_server_number() or spapr_vcpu_id().
2658 spapr_set_vsmt_mode(spapr
, &error_fatal
);
2660 /* Set up Interrupt Controller before we create the VCPUs */
2661 spapr_irq_init(spapr
, &error_fatal
);
2663 /* Set up containers for ibm,client-architecture-support negotiated options
2665 spapr
->ov5
= spapr_ovec_new();
2666 spapr
->ov5_cas
= spapr_ovec_new();
2668 if (smc
->dr_lmb_enabled
) {
2669 spapr_ovec_set(spapr
->ov5
, OV5_DRCONF_MEMORY
);
2670 spapr_validate_node_memory(machine
, &error_fatal
);
2673 spapr_ovec_set(spapr
->ov5
, OV5_FORM1_AFFINITY
);
2675 /* advertise support for dedicated HP event source to guests */
2676 if (spapr
->use_hotplug_event_source
) {
2677 spapr_ovec_set(spapr
->ov5
, OV5_HP_EVT
);
2680 /* advertise support for HPT resizing */
2681 if (spapr
->resize_hpt
!= SPAPR_RESIZE_HPT_DISABLED
) {
2682 spapr_ovec_set(spapr
->ov5
, OV5_HPT_RESIZE
);
2685 /* advertise support for ibm,dyamic-memory-v2 */
2686 spapr_ovec_set(spapr
->ov5
, OV5_DRMEM_V2
);
2688 /* advertise XIVE on POWER9 machines */
2689 if (spapr
->irq
->ov5
& (SPAPR_OV5_XIVE_EXPLOIT
| SPAPR_OV5_XIVE_BOTH
)) {
2690 if (ppc_type_check_compat(machine
->cpu_type
, CPU_POWERPC_LOGICAL_3_00
,
2691 0, spapr
->max_compat_pvr
)) {
2692 spapr_ovec_set(spapr
->ov5
, OV5_XIVE_EXPLOIT
);
2693 } else if (spapr
->irq
->ov5
& SPAPR_OV5_XIVE_EXPLOIT
) {
2694 error_report("XIVE-only machines require a POWER9 CPU");
2700 spapr_init_cpus(spapr
);
2702 if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2703 ppc_type_check_compat(machine
->cpu_type
, CPU_POWERPC_LOGICAL_3_00
, 0,
2704 spapr
->max_compat_pvr
)) {
2705 /* KVM and TCG always allow GTSE with radix... */
2706 spapr_ovec_set(spapr
->ov5
, OV5_MMU_RADIX_GTSE
);
2708 /* ... but not with hash (currently). */
2710 if (kvm_enabled()) {
2711 /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2712 kvmppc_enable_logical_ci_hcalls();
2713 kvmppc_enable_set_mode_hcall();
2715 /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2716 kvmppc_enable_clear_ref_mod_hcalls();
2720 memory_region_allocate_system_memory(ram
, NULL
, "ppc_spapr.ram",
2722 memory_region_add_subregion(sysmem
, 0, ram
);
2724 /* always allocate the device memory information */
2725 machine
->device_memory
= g_malloc0(sizeof(*machine
->device_memory
));
2727 /* initialize hotplug memory address space */
2728 if (machine
->ram_size
< machine
->maxram_size
) {
2729 ram_addr_t device_mem_size
= machine
->maxram_size
- machine
->ram_size
;
2731 * Limit the number of hotpluggable memory slots to half the number
2732 * slots that KVM supports, leaving the other half for PCI and other
2733 * devices. However ensure that number of slots doesn't drop below 32.
2735 int max_memslots
= kvm_enabled() ? kvm_get_max_memslots() / 2 :
2736 SPAPR_MAX_RAM_SLOTS
;
2738 if (max_memslots
< SPAPR_MAX_RAM_SLOTS
) {
2739 max_memslots
= SPAPR_MAX_RAM_SLOTS
;
2741 if (machine
->ram_slots
> max_memslots
) {
2742 error_report("Specified number of memory slots %"
2743 PRIu64
" exceeds max supported %d",
2744 machine
->ram_slots
, max_memslots
);
2748 machine
->device_memory
->base
= ROUND_UP(machine
->ram_size
,
2749 SPAPR_DEVICE_MEM_ALIGN
);
2750 memory_region_init(&machine
->device_memory
->mr
, OBJECT(spapr
),
2751 "device-memory", device_mem_size
);
2752 memory_region_add_subregion(sysmem
, machine
->device_memory
->base
,
2753 &machine
->device_memory
->mr
);
2756 if (smc
->dr_lmb_enabled
) {
2757 spapr_create_lmb_dr_connectors(spapr
);
2760 filename
= qemu_find_file(QEMU_FILE_TYPE_BIOS
, "spapr-rtas.bin");
2762 error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2765 spapr
->rtas_size
= get_image_size(filename
);
2766 if (spapr
->rtas_size
< 0) {
2767 error_report("Could not get size of LPAR rtas '%s'", filename
);
2770 spapr
->rtas_blob
= g_malloc(spapr
->rtas_size
);
2771 if (load_image_size(filename
, spapr
->rtas_blob
, spapr
->rtas_size
) < 0) {
2772 error_report("Could not load LPAR rtas '%s'", filename
);
2775 if (spapr
->rtas_size
> RTAS_MAX_SIZE
) {
2776 error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2777 (size_t)spapr
->rtas_size
, RTAS_MAX_SIZE
);
2782 /* Set up RTAS event infrastructure */
2783 spapr_events_init(spapr
);
2785 /* Set up the RTC RTAS interfaces */
2786 spapr_rtc_create(spapr
);
2788 /* Set up VIO bus */
2789 spapr
->vio_bus
= spapr_vio_bus_init();
2791 for (i
= 0; i
< serial_max_hds(); i
++) {
2793 spapr_vty_create(spapr
->vio_bus
, serial_hd(i
));
2797 /* We always have at least the nvram device on VIO */
2798 spapr_create_nvram(spapr
);
2801 spapr_pci_rtas_init();
2803 phb
= spapr_create_default_phb();
2805 for (i
= 0; i
< nb_nics
; i
++) {
2806 NICInfo
*nd
= &nd_table
[i
];
2809 nd
->model
= g_strdup("spapr-vlan");
2812 if (g_str_equal(nd
->model
, "spapr-vlan") ||
2813 g_str_equal(nd
->model
, "ibmveth")) {
2814 spapr_vlan_create(spapr
->vio_bus
, nd
);
2816 pci_nic_init_nofail(&nd_table
[i
], phb
->bus
, nd
->model
, NULL
);
2820 for (i
= 0; i
<= drive_get_max_bus(IF_SCSI
); i
++) {
2821 spapr_vscsi_create(spapr
->vio_bus
);
2825 if (spapr_vga_init(phb
->bus
, &error_fatal
)) {
2826 spapr
->has_graphics
= true;
2827 machine
->usb
|= defaults_enabled() && !machine
->usb_disabled
;
2831 if (smc
->use_ohci_by_default
) {
2832 pci_create_simple(phb
->bus
, -1, "pci-ohci");
2834 pci_create_simple(phb
->bus
, -1, "nec-usb-xhci");
2837 if (spapr
->has_graphics
) {
2838 USBBus
*usb_bus
= usb_bus_find(-1);
2840 usb_create_simple(usb_bus
, "usb-kbd");
2841 usb_create_simple(usb_bus
, "usb-mouse");
2845 if (spapr
->rma_size
< (MIN_RMA_SLOF
* MiB
)) {
2847 "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2852 if (kernel_filename
) {
2853 uint64_t lowaddr
= 0;
2855 spapr
->kernel_size
= load_elf(kernel_filename
, translate_kernel_address
,
2856 NULL
, NULL
, &lowaddr
, NULL
, 1,
2857 PPC_ELF_MACHINE
, 0, 0);
2858 if (spapr
->kernel_size
== ELF_LOAD_WRONG_ENDIAN
) {
2859 spapr
->kernel_size
= load_elf(kernel_filename
,
2860 translate_kernel_address
, NULL
, NULL
,
2861 &lowaddr
, NULL
, 0, PPC_ELF_MACHINE
,
2863 spapr
->kernel_le
= spapr
->kernel_size
> 0;
2865 if (spapr
->kernel_size
< 0) {
2866 error_report("error loading %s: %s", kernel_filename
,
2867 load_elf_strerror(spapr
->kernel_size
));
2872 if (initrd_filename
) {
2873 /* Try to locate the initrd in the gap between the kernel
2874 * and the firmware. Add a bit of space just in case
2876 spapr
->initrd_base
= (KERNEL_LOAD_ADDR
+ spapr
->kernel_size
2877 + 0x1ffff) & ~0xffff;
2878 spapr
->initrd_size
= load_image_targphys(initrd_filename
,
2881 - spapr
->initrd_base
);
2882 if (spapr
->initrd_size
< 0) {
2883 error_report("could not load initial ram disk '%s'",
2890 if (bios_name
== NULL
) {
2891 bios_name
= FW_FILE_NAME
;
2893 filename
= qemu_find_file(QEMU_FILE_TYPE_BIOS
, bios_name
);
2895 error_report("Could not find LPAR firmware '%s'", bios_name
);
2898 fw_size
= load_image_targphys(filename
, 0, FW_MAX_SIZE
);
2900 error_report("Could not load LPAR firmware '%s'", filename
);
2905 /* FIXME: Should register things through the MachineState's qdev
2906 * interface, this is a legacy from the sPAPREnvironment structure
2907 * which predated MachineState but had a similar function */
2908 vmstate_register(NULL
, 0, &vmstate_spapr
, spapr
);
2909 register_savevm_live(NULL
, "spapr/htab", -1, 1,
2910 &savevm_htab_handlers
, spapr
);
2912 qemu_register_boot_set(spapr_boot_set
, spapr
);
2914 if (kvm_enabled()) {
2915 /* to stop and start vmclock */
2916 qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change
,
2919 kvmppc_spapr_enable_inkernel_multitce();
2923 static int spapr_kvm_type(const char *vm_type
)
2929 if (!strcmp(vm_type
, "HV")) {
2933 if (!strcmp(vm_type
, "PR")) {
2937 error_report("Unknown kvm-type specified '%s'", vm_type
);
2942 * Implementation of an interface to adjust firmware path
2943 * for the bootindex property handling.
2945 static char *spapr_get_fw_dev_path(FWPathProvider
*p
, BusState
*bus
,
2948 #define CAST(type, obj, name) \
2949 ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2950 SCSIDevice
*d
= CAST(SCSIDevice
, dev
, TYPE_SCSI_DEVICE
);
2951 sPAPRPHBState
*phb
= CAST(sPAPRPHBState
, dev
, TYPE_SPAPR_PCI_HOST_BRIDGE
);
2952 VHostSCSICommon
*vsc
= CAST(VHostSCSICommon
, dev
, TYPE_VHOST_SCSI_COMMON
);
2955 void *spapr
= CAST(void, bus
->parent
, "spapr-vscsi");
2956 VirtIOSCSI
*virtio
= CAST(VirtIOSCSI
, bus
->parent
, TYPE_VIRTIO_SCSI
);
2957 USBDevice
*usb
= CAST(USBDevice
, bus
->parent
, TYPE_USB_DEVICE
);
2961 * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2962 * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2963 * in the top 16 bits of the 64-bit LUN
2965 unsigned id
= 0x8000 | (d
->id
<< 8) | d
->lun
;
2966 return g_strdup_printf("%s@%"PRIX64
, qdev_fw_name(dev
),
2967 (uint64_t)id
<< 48);
2968 } else if (virtio
) {
2970 * We use SRP luns of the form 01000000 | (target << 8) | lun
2971 * in the top 32 bits of the 64-bit LUN
2972 * Note: the quote above is from SLOF and it is wrong,
2973 * the actual binding is:
2974 * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2976 unsigned id
= 0x1000000 | (d
->id
<< 16) | d
->lun
;
2977 if (d
->lun
>= 256) {
2978 /* Use the LUN "flat space addressing method" */
2981 return g_strdup_printf("%s@%"PRIX64
, qdev_fw_name(dev
),
2982 (uint64_t)id
<< 32);
2985 * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2986 * in the top 32 bits of the 64-bit LUN
2988 unsigned usb_port
= atoi(usb
->port
->path
);
2989 unsigned id
= 0x1000000 | (usb_port
<< 16) | d
->lun
;
2990 return g_strdup_printf("%s@%"PRIX64
, qdev_fw_name(dev
),
2991 (uint64_t)id
<< 32);
2996 * SLOF probes the USB devices, and if it recognizes that the device is a
2997 * storage device, it changes its name to "storage" instead of "usb-host",
2998 * and additionally adds a child node for the SCSI LUN, so the correct
2999 * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3001 if (strcmp("usb-host", qdev_fw_name(dev
)) == 0) {
3002 USBDevice
*usbdev
= CAST(USBDevice
, dev
, TYPE_USB_DEVICE
);
3003 if (usb_host_dev_is_scsi_storage(usbdev
)) {
3004 return g_strdup_printf("storage@%s/disk", usbdev
->port
->path
);
3009 /* Replace "pci" with "pci@800000020000000" */
3010 return g_strdup_printf("pci@%"PRIX64
, phb
->buid
);
3014 /* Same logic as virtio above */
3015 unsigned id
= 0x1000000 | (vsc
->target
<< 16) | vsc
->lun
;
3016 return g_strdup_printf("disk@%"PRIX64
, (uint64_t)id
<< 32);
3019 if (g_str_equal("pci-bridge", qdev_fw_name(dev
))) {
3020 /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3021 PCIDevice
*pcidev
= CAST(PCIDevice
, dev
, TYPE_PCI_DEVICE
);
3022 return g_strdup_printf("pci@%x", PCI_SLOT(pcidev
->devfn
));
3028 static char *spapr_get_kvm_type(Object
*obj
, Error
**errp
)
3030 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
3032 return g_strdup(spapr
->kvm_type
);
3035 static void spapr_set_kvm_type(Object
*obj
, const char *value
, Error
**errp
)
3037 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
3039 g_free(spapr
->kvm_type
);
3040 spapr
->kvm_type
= g_strdup(value
);
3043 static bool spapr_get_modern_hotplug_events(Object
*obj
, Error
**errp
)
3045 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
3047 return spapr
->use_hotplug_event_source
;
3050 static void spapr_set_modern_hotplug_events(Object
*obj
, bool value
,
3053 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
3055 spapr
->use_hotplug_event_source
= value
;
3058 static bool spapr_get_msix_emulation(Object
*obj
, Error
**errp
)
3063 static char *spapr_get_resize_hpt(Object
*obj
, Error
**errp
)
3065 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
3067 switch (spapr
->resize_hpt
) {
3068 case SPAPR_RESIZE_HPT_DEFAULT
:
3069 return g_strdup("default");
3070 case SPAPR_RESIZE_HPT_DISABLED
:
3071 return g_strdup("disabled");
3072 case SPAPR_RESIZE_HPT_ENABLED
:
3073 return g_strdup("enabled");
3074 case SPAPR_RESIZE_HPT_REQUIRED
:
3075 return g_strdup("required");
3077 g_assert_not_reached();
3080 static void spapr_set_resize_hpt(Object
*obj
, const char *value
, Error
**errp
)
3082 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
3084 if (strcmp(value
, "default") == 0) {
3085 spapr
->resize_hpt
= SPAPR_RESIZE_HPT_DEFAULT
;
3086 } else if (strcmp(value
, "disabled") == 0) {
3087 spapr
->resize_hpt
= SPAPR_RESIZE_HPT_DISABLED
;
3088 } else if (strcmp(value
, "enabled") == 0) {
3089 spapr
->resize_hpt
= SPAPR_RESIZE_HPT_ENABLED
;
3090 } else if (strcmp(value
, "required") == 0) {
3091 spapr
->resize_hpt
= SPAPR_RESIZE_HPT_REQUIRED
;
3093 error_setg(errp
, "Bad value for \"resize-hpt\" property");
3097 static void spapr_get_vsmt(Object
*obj
, Visitor
*v
, const char *name
,
3098 void *opaque
, Error
**errp
)
3100 visit_type_uint32(v
, name
, (uint32_t *)opaque
, errp
);
3103 static void spapr_set_vsmt(Object
*obj
, Visitor
*v
, const char *name
,
3104 void *opaque
, Error
**errp
)
3106 visit_type_uint32(v
, name
, (uint32_t *)opaque
, errp
);
3109 static char *spapr_get_ic_mode(Object
*obj
, Error
**errp
)
3111 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
3113 if (spapr
->irq
== &spapr_irq_xics_legacy
) {
3114 return g_strdup("legacy");
3115 } else if (spapr
->irq
== &spapr_irq_xics
) {
3116 return g_strdup("xics");
3117 } else if (spapr
->irq
== &spapr_irq_xive
) {
3118 return g_strdup("xive");
3119 } else if (spapr
->irq
== &spapr_irq_dual
) {
3120 return g_strdup("dual");
3122 g_assert_not_reached();
3125 static void spapr_set_ic_mode(Object
*obj
, const char *value
, Error
**errp
)
3127 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
3129 /* The legacy IRQ backend can not be set */
3130 if (strcmp(value
, "xics") == 0) {
3131 spapr
->irq
= &spapr_irq_xics
;
3132 } else if (strcmp(value
, "xive") == 0) {
3133 spapr
->irq
= &spapr_irq_xive
;
3134 } else if (strcmp(value
, "dual") == 0) {
3135 spapr
->irq
= &spapr_irq_dual
;
3137 error_setg(errp
, "Bad value for \"ic-mode\" property");
3141 static void spapr_instance_init(Object
*obj
)
3143 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
3144 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(spapr
);
3146 spapr
->htab_fd
= -1;
3147 spapr
->use_hotplug_event_source
= true;
3148 object_property_add_str(obj
, "kvm-type",
3149 spapr_get_kvm_type
, spapr_set_kvm_type
, NULL
);
3150 object_property_set_description(obj
, "kvm-type",
3151 "Specifies the KVM virtualization mode (HV, PR)",
3153 object_property_add_bool(obj
, "modern-hotplug-events",
3154 spapr_get_modern_hotplug_events
,
3155 spapr_set_modern_hotplug_events
,
3157 object_property_set_description(obj
, "modern-hotplug-events",
3158 "Use dedicated hotplug event mechanism in"
3159 " place of standard EPOW events when possible"
3160 " (required for memory hot-unplug support)",
3162 ppc_compat_add_property(obj
, "max-cpu-compat", &spapr
->max_compat_pvr
,
3163 "Maximum permitted CPU compatibility mode",
3166 object_property_add_str(obj
, "resize-hpt",
3167 spapr_get_resize_hpt
, spapr_set_resize_hpt
, NULL
);
3168 object_property_set_description(obj
, "resize-hpt",
3169 "Resizing of the Hash Page Table (enabled, disabled, required)",
3171 object_property_add(obj
, "vsmt", "uint32", spapr_get_vsmt
,
3172 spapr_set_vsmt
, NULL
, &spapr
->vsmt
, &error_abort
);
3173 object_property_set_description(obj
, "vsmt",
3174 "Virtual SMT: KVM behaves as if this were"
3175 " the host's SMT mode", &error_abort
);
3176 object_property_add_bool(obj
, "vfio-no-msix-emulation",
3177 spapr_get_msix_emulation
, NULL
, NULL
);
3179 /* The machine class defines the default interrupt controller mode */
3180 spapr
->irq
= smc
->irq
;
3181 object_property_add_str(obj
, "ic-mode", spapr_get_ic_mode
,
3182 spapr_set_ic_mode
, NULL
);
3183 object_property_set_description(obj
, "ic-mode",
3184 "Specifies the interrupt controller mode (xics, xive, dual)",
3188 static void spapr_machine_finalizefn(Object
*obj
)
3190 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
3192 g_free(spapr
->kvm_type
);
3195 void spapr_do_system_reset_on_cpu(CPUState
*cs
, run_on_cpu_data arg
)
3197 cpu_synchronize_state(cs
);
3198 ppc_cpu_do_system_reset(cs
);
3201 static void spapr_nmi(NMIState
*n
, int cpu_index
, Error
**errp
)
3206 async_run_on_cpu(cs
, spapr_do_system_reset_on_cpu
, RUN_ON_CPU_NULL
);
3210 static void spapr_add_lmbs(DeviceState
*dev
, uint64_t addr_start
, uint64_t size
,
3211 uint32_t node
, bool dedicated_hp_event_source
,
3214 sPAPRDRConnector
*drc
;
3215 uint32_t nr_lmbs
= size
/SPAPR_MEMORY_BLOCK_SIZE
;
3216 int i
, fdt_offset
, fdt_size
;
3218 uint64_t addr
= addr_start
;
3219 bool hotplugged
= spapr_drc_hotplugged(dev
);
3220 Error
*local_err
= NULL
;
3222 for (i
= 0; i
< nr_lmbs
; i
++) {
3223 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
3224 addr
/ SPAPR_MEMORY_BLOCK_SIZE
);
3227 fdt
= create_device_tree(&fdt_size
);
3228 fdt_offset
= spapr_populate_memory_node(fdt
, node
, addr
,
3229 SPAPR_MEMORY_BLOCK_SIZE
);
3231 spapr_drc_attach(drc
, dev
, fdt
, fdt_offset
, &local_err
);
3233 while (addr
> addr_start
) {
3234 addr
-= SPAPR_MEMORY_BLOCK_SIZE
;
3235 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
3236 addr
/ SPAPR_MEMORY_BLOCK_SIZE
);
3237 spapr_drc_detach(drc
);
3240 error_propagate(errp
, local_err
);
3244 spapr_drc_reset(drc
);
3246 addr
+= SPAPR_MEMORY_BLOCK_SIZE
;
3248 /* send hotplug notification to the
3249 * guest only in case of hotplugged memory
3252 if (dedicated_hp_event_source
) {
3253 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
3254 addr_start
/ SPAPR_MEMORY_BLOCK_SIZE
);
3255 spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB
,
3257 spapr_drc_index(drc
));
3259 spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB
,
3265 static void spapr_memory_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
3268 Error
*local_err
= NULL
;
3269 sPAPRMachineState
*ms
= SPAPR_MACHINE(hotplug_dev
);
3270 PCDIMMDevice
*dimm
= PC_DIMM(dev
);
3271 uint64_t size
, addr
;
3274 size
= memory_device_get_region_size(MEMORY_DEVICE(dev
), &error_abort
);
3276 pc_dimm_plug(dimm
, MACHINE(ms
), &local_err
);
3281 addr
= object_property_get_uint(OBJECT(dimm
),
3282 PC_DIMM_ADDR_PROP
, &local_err
);
3287 node
= object_property_get_uint(OBJECT(dev
), PC_DIMM_NODE_PROP
,
3289 spapr_add_lmbs(dev
, addr
, size
, node
,
3290 spapr_ovec_test(ms
->ov5_cas
, OV5_HP_EVT
),
3299 pc_dimm_unplug(dimm
, MACHINE(ms
));
3301 error_propagate(errp
, local_err
);
3304 static void spapr_memory_pre_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
3307 const sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(hotplug_dev
);
3308 sPAPRMachineState
*spapr
= SPAPR_MACHINE(hotplug_dev
);
3309 PCDIMMDevice
*dimm
= PC_DIMM(dev
);
3310 Error
*local_err
= NULL
;
3315 if (!smc
->dr_lmb_enabled
) {
3316 error_setg(errp
, "Memory hotplug not supported for this machine");
3320 size
= memory_device_get_region_size(MEMORY_DEVICE(dimm
), &local_err
);
3322 error_propagate(errp
, local_err
);
3326 if (size
% SPAPR_MEMORY_BLOCK_SIZE
) {
3327 error_setg(errp
, "Hotplugged memory size must be a multiple of "
3328 "%" PRIu64
" MB", SPAPR_MEMORY_BLOCK_SIZE
/ MiB
);
3332 memdev
= object_property_get_link(OBJECT(dimm
), PC_DIMM_MEMDEV_PROP
,
3334 pagesize
= host_memory_backend_pagesize(MEMORY_BACKEND(memdev
));
3335 spapr_check_pagesize(spapr
, pagesize
, &local_err
);
3337 error_propagate(errp
, local_err
);
3341 pc_dimm_pre_plug(dimm
, MACHINE(hotplug_dev
), NULL
, errp
);
3344 struct sPAPRDIMMState
{
3347 QTAILQ_ENTRY(sPAPRDIMMState
) next
;
3350 static sPAPRDIMMState
*spapr_pending_dimm_unplugs_find(sPAPRMachineState
*s
,
3353 sPAPRDIMMState
*dimm_state
= NULL
;
3355 QTAILQ_FOREACH(dimm_state
, &s
->pending_dimm_unplugs
, next
) {
3356 if (dimm_state
->dimm
== dimm
) {
3363 static sPAPRDIMMState
*spapr_pending_dimm_unplugs_add(sPAPRMachineState
*spapr
,
3367 sPAPRDIMMState
*ds
= NULL
;
3370 * If this request is for a DIMM whose removal had failed earlier
3371 * (due to guest's refusal to remove the LMBs), we would have this
3372 * dimm already in the pending_dimm_unplugs list. In that
3373 * case don't add again.
3375 ds
= spapr_pending_dimm_unplugs_find(spapr
, dimm
);
3377 ds
= g_malloc0(sizeof(sPAPRDIMMState
));
3378 ds
->nr_lmbs
= nr_lmbs
;
3380 QTAILQ_INSERT_HEAD(&spapr
->pending_dimm_unplugs
, ds
, next
);
3385 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState
*spapr
,
3386 sPAPRDIMMState
*dimm_state
)
3388 QTAILQ_REMOVE(&spapr
->pending_dimm_unplugs
, dimm_state
, next
);
3392 static sPAPRDIMMState
*spapr_recover_pending_dimm_state(sPAPRMachineState
*ms
,
3395 sPAPRDRConnector
*drc
;
3396 uint64_t size
= memory_device_get_region_size(MEMORY_DEVICE(dimm
),
3398 uint32_t nr_lmbs
= size
/ SPAPR_MEMORY_BLOCK_SIZE
;
3399 uint32_t avail_lmbs
= 0;
3400 uint64_t addr_start
, addr
;
3403 addr_start
= object_property_get_int(OBJECT(dimm
), PC_DIMM_ADDR_PROP
,
3407 for (i
= 0; i
< nr_lmbs
; i
++) {
3408 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
3409 addr
/ SPAPR_MEMORY_BLOCK_SIZE
);
3414 addr
+= SPAPR_MEMORY_BLOCK_SIZE
;
3417 return spapr_pending_dimm_unplugs_add(ms
, avail_lmbs
, dimm
);
3420 /* Callback to be called during DRC release. */
3421 void spapr_lmb_release(DeviceState
*dev
)
3423 HotplugHandler
*hotplug_ctrl
= qdev_get_hotplug_handler(dev
);
3424 sPAPRMachineState
*spapr
= SPAPR_MACHINE(hotplug_ctrl
);
3425 sPAPRDIMMState
*ds
= spapr_pending_dimm_unplugs_find(spapr
, PC_DIMM(dev
));
3427 /* This information will get lost if a migration occurs
3428 * during the unplug process. In this case recover it. */
3430 ds
= spapr_recover_pending_dimm_state(spapr
, PC_DIMM(dev
));
3432 /* The DRC being examined by the caller at least must be counted */
3433 g_assert(ds
->nr_lmbs
);
3436 if (--ds
->nr_lmbs
) {
3441 * Now that all the LMBs have been removed by the guest, call the
3442 * unplug handler chain. This can never fail.
3444 hotplug_handler_unplug(hotplug_ctrl
, dev
, &error_abort
);
3447 static void spapr_memory_unplug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
)
3449 sPAPRMachineState
*spapr
= SPAPR_MACHINE(hotplug_dev
);
3450 sPAPRDIMMState
*ds
= spapr_pending_dimm_unplugs_find(spapr
, PC_DIMM(dev
));
3452 pc_dimm_unplug(PC_DIMM(dev
), MACHINE(hotplug_dev
));
3453 object_unparent(OBJECT(dev
));
3454 spapr_pending_dimm_unplugs_remove(spapr
, ds
);
3457 static void spapr_memory_unplug_request(HotplugHandler
*hotplug_dev
,
3458 DeviceState
*dev
, Error
**errp
)
3460 sPAPRMachineState
*spapr
= SPAPR_MACHINE(hotplug_dev
);
3461 Error
*local_err
= NULL
;
3462 PCDIMMDevice
*dimm
= PC_DIMM(dev
);
3464 uint64_t size
, addr_start
, addr
;
3466 sPAPRDRConnector
*drc
;
3468 size
= memory_device_get_region_size(MEMORY_DEVICE(dimm
), &error_abort
);
3469 nr_lmbs
= size
/ SPAPR_MEMORY_BLOCK_SIZE
;
3471 addr_start
= object_property_get_uint(OBJECT(dimm
), PC_DIMM_ADDR_PROP
,
3478 * An existing pending dimm state for this DIMM means that there is an
3479 * unplug operation in progress, waiting for the spapr_lmb_release
3480 * callback to complete the job (BQL can't cover that far). In this case,
3481 * bail out to avoid detaching DRCs that were already released.
3483 if (spapr_pending_dimm_unplugs_find(spapr
, dimm
)) {
3484 error_setg(&local_err
,
3485 "Memory unplug already in progress for device %s",
3490 spapr_pending_dimm_unplugs_add(spapr
, nr_lmbs
, dimm
);
3493 for (i
= 0; i
< nr_lmbs
; i
++) {
3494 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
3495 addr
/ SPAPR_MEMORY_BLOCK_SIZE
);
3498 spapr_drc_detach(drc
);
3499 addr
+= SPAPR_MEMORY_BLOCK_SIZE
;
3502 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
3503 addr_start
/ SPAPR_MEMORY_BLOCK_SIZE
);
3504 spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB
,
3505 nr_lmbs
, spapr_drc_index(drc
));
3507 error_propagate(errp
, local_err
);
3510 static void *spapr_populate_hotplug_cpu_dt(CPUState
*cs
, int *fdt_offset
,
3511 sPAPRMachineState
*spapr
)
3513 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
3514 DeviceClass
*dc
= DEVICE_GET_CLASS(cs
);
3515 int id
= spapr_get_vcpu_id(cpu
);
3517 int offset
, fdt_size
;
3520 fdt
= create_device_tree(&fdt_size
);
3521 nodename
= g_strdup_printf("%s@%x", dc
->fw_name
, id
);
3522 offset
= fdt_add_subnode(fdt
, 0, nodename
);
3524 spapr_populate_cpu_dt(cs
, fdt
, offset
, spapr
);
3527 *fdt_offset
= offset
;
3531 /* Callback to be called during DRC release. */
3532 void spapr_core_release(DeviceState
*dev
)
3534 HotplugHandler
*hotplug_ctrl
= qdev_get_hotplug_handler(dev
);
3536 /* Call the unplug handler chain. This can never fail. */
3537 hotplug_handler_unplug(hotplug_ctrl
, dev
, &error_abort
);
3540 static void spapr_core_unplug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
)
3542 MachineState
*ms
= MACHINE(hotplug_dev
);
3543 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(ms
);
3544 CPUCore
*cc
= CPU_CORE(dev
);
3545 CPUArchId
*core_slot
= spapr_find_cpu_slot(ms
, cc
->core_id
, NULL
);
3547 if (smc
->pre_2_10_has_unused_icps
) {
3548 sPAPRCPUCore
*sc
= SPAPR_CPU_CORE(OBJECT(dev
));
3551 for (i
= 0; i
< cc
->nr_threads
; i
++) {
3552 CPUState
*cs
= CPU(sc
->threads
[i
]);
3554 pre_2_10_vmstate_register_dummy_icp(cs
->cpu_index
);
3559 core_slot
->cpu
= NULL
;
3560 object_unparent(OBJECT(dev
));
3564 void spapr_core_unplug_request(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
3567 sPAPRMachineState
*spapr
= SPAPR_MACHINE(OBJECT(hotplug_dev
));
3569 sPAPRDRConnector
*drc
;
3570 CPUCore
*cc
= CPU_CORE(dev
);
3572 if (!spapr_find_cpu_slot(MACHINE(hotplug_dev
), cc
->core_id
, &index
)) {
3573 error_setg(errp
, "Unable to find CPU core with core-id: %d",
3578 error_setg(errp
, "Boot CPU core may not be unplugged");
3582 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_CPU
,
3583 spapr_vcpu_id(spapr
, cc
->core_id
));
3586 spapr_drc_detach(drc
);
3588 spapr_hotplug_req_remove_by_index(drc
);
3591 static void spapr_core_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
3594 sPAPRMachineState
*spapr
= SPAPR_MACHINE(OBJECT(hotplug_dev
));
3595 MachineClass
*mc
= MACHINE_GET_CLASS(spapr
);
3596 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
3597 sPAPRCPUCore
*core
= SPAPR_CPU_CORE(OBJECT(dev
));
3598 CPUCore
*cc
= CPU_CORE(dev
);
3599 CPUState
*cs
= CPU(core
->threads
[0]);
3600 sPAPRDRConnector
*drc
;
3601 Error
*local_err
= NULL
;
3602 CPUArchId
*core_slot
;
3604 bool hotplugged
= spapr_drc_hotplugged(dev
);
3606 core_slot
= spapr_find_cpu_slot(MACHINE(hotplug_dev
), cc
->core_id
, &index
);
3608 error_setg(errp
, "Unable to find CPU core with core-id: %d",
3612 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_CPU
,
3613 spapr_vcpu_id(spapr
, cc
->core_id
));
3615 g_assert(drc
|| !mc
->has_hotpluggable_cpus
);
3621 fdt
= spapr_populate_hotplug_cpu_dt(cs
, &fdt_offset
, spapr
);
3623 spapr_drc_attach(drc
, dev
, fdt
, fdt_offset
, &local_err
);
3626 error_propagate(errp
, local_err
);
3632 * Send hotplug notification interrupt to the guest only
3633 * in case of hotplugged CPUs.
3635 spapr_hotplug_req_add_by_index(drc
);
3637 spapr_drc_reset(drc
);
3641 core_slot
->cpu
= OBJECT(dev
);
3643 if (smc
->pre_2_10_has_unused_icps
) {
3646 for (i
= 0; i
< cc
->nr_threads
; i
++) {
3647 cs
= CPU(core
->threads
[i
]);
3648 pre_2_10_vmstate_unregister_dummy_icp(cs
->cpu_index
);
3653 static void spapr_core_pre_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
3656 MachineState
*machine
= MACHINE(OBJECT(hotplug_dev
));
3657 MachineClass
*mc
= MACHINE_GET_CLASS(hotplug_dev
);
3658 Error
*local_err
= NULL
;
3659 CPUCore
*cc
= CPU_CORE(dev
);
3660 const char *base_core_type
= spapr_get_cpu_core_type(machine
->cpu_type
);
3661 const char *type
= object_get_typename(OBJECT(dev
));
3662 CPUArchId
*core_slot
;
3665 if (dev
->hotplugged
&& !mc
->has_hotpluggable_cpus
) {
3666 error_setg(&local_err
, "CPU hotplug not supported for this machine");
3670 if (strcmp(base_core_type
, type
)) {
3671 error_setg(&local_err
, "CPU core type should be %s", base_core_type
);
3675 if (cc
->core_id
% smp_threads
) {
3676 error_setg(&local_err
, "invalid core id %d", cc
->core_id
);
3681 * In general we should have homogeneous threads-per-core, but old
3682 * (pre hotplug support) machine types allow the last core to have
3683 * reduced threads as a compatibility hack for when we allowed
3684 * total vcpus not a multiple of threads-per-core.
3686 if (mc
->has_hotpluggable_cpus
&& (cc
->nr_threads
!= smp_threads
)) {
3687 error_setg(&local_err
, "invalid nr-threads %d, must be %d",
3688 cc
->nr_threads
, smp_threads
);
3692 core_slot
= spapr_find_cpu_slot(MACHINE(hotplug_dev
), cc
->core_id
, &index
);
3694 error_setg(&local_err
, "core id %d out of range", cc
->core_id
);
3698 if (core_slot
->cpu
) {
3699 error_setg(&local_err
, "core %d already populated", cc
->core_id
);
3703 numa_cpu_pre_plug(core_slot
, dev
, &local_err
);
3706 error_propagate(errp
, local_err
);
3709 static void spapr_machine_device_plug(HotplugHandler
*hotplug_dev
,
3710 DeviceState
*dev
, Error
**errp
)
3712 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
)) {
3713 spapr_memory_plug(hotplug_dev
, dev
, errp
);
3714 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
3715 spapr_core_plug(hotplug_dev
, dev
, errp
);
3719 static void spapr_machine_device_unplug(HotplugHandler
*hotplug_dev
,
3720 DeviceState
*dev
, Error
**errp
)
3722 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
)) {
3723 spapr_memory_unplug(hotplug_dev
, dev
);
3724 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
3725 spapr_core_unplug(hotplug_dev
, dev
);
3729 static void spapr_machine_device_unplug_request(HotplugHandler
*hotplug_dev
,
3730 DeviceState
*dev
, Error
**errp
)
3732 sPAPRMachineState
*sms
= SPAPR_MACHINE(OBJECT(hotplug_dev
));
3733 MachineClass
*mc
= MACHINE_GET_CLASS(sms
);
3735 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
)) {
3736 if (spapr_ovec_test(sms
->ov5_cas
, OV5_HP_EVT
)) {
3737 spapr_memory_unplug_request(hotplug_dev
, dev
, errp
);
3739 /* NOTE: this means there is a window after guest reset, prior to
3740 * CAS negotiation, where unplug requests will fail due to the
3741 * capability not being detected yet. This is a bit different than
3742 * the case with PCI unplug, where the events will be queued and
3743 * eventually handled by the guest after boot
3745 error_setg(errp
, "Memory hot unplug not supported for this guest");
3747 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
3748 if (!mc
->has_hotpluggable_cpus
) {
3749 error_setg(errp
, "CPU hot unplug not supported on this machine");
3752 spapr_core_unplug_request(hotplug_dev
, dev
, errp
);
3756 static void spapr_machine_device_pre_plug(HotplugHandler
*hotplug_dev
,
3757 DeviceState
*dev
, Error
**errp
)
3759 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
)) {
3760 spapr_memory_pre_plug(hotplug_dev
, dev
, errp
);
3761 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
3762 spapr_core_pre_plug(hotplug_dev
, dev
, errp
);
3766 static HotplugHandler
*spapr_get_hotplug_handler(MachineState
*machine
,
3769 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
) ||
3770 object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
3771 return HOTPLUG_HANDLER(machine
);
3776 static CpuInstanceProperties
3777 spapr_cpu_index_to_props(MachineState
*machine
, unsigned cpu_index
)
3779 CPUArchId
*core_slot
;
3780 MachineClass
*mc
= MACHINE_GET_CLASS(machine
);
3782 /* make sure possible_cpu are intialized */
3783 mc
->possible_cpu_arch_ids(machine
);
3784 /* get CPU core slot containing thread that matches cpu_index */
3785 core_slot
= spapr_find_cpu_slot(machine
, cpu_index
, NULL
);
3787 return core_slot
->props
;
3790 static int64_t spapr_get_default_cpu_node_id(const MachineState
*ms
, int idx
)
3792 return idx
/ smp_cores
% nb_numa_nodes
;
3795 static const CPUArchIdList
*spapr_possible_cpu_arch_ids(MachineState
*machine
)
3798 const char *core_type
;
3799 int spapr_max_cores
= max_cpus
/ smp_threads
;
3800 MachineClass
*mc
= MACHINE_GET_CLASS(machine
);
3802 if (!mc
->has_hotpluggable_cpus
) {
3803 spapr_max_cores
= QEMU_ALIGN_UP(smp_cpus
, smp_threads
) / smp_threads
;
3805 if (machine
->possible_cpus
) {
3806 assert(machine
->possible_cpus
->len
== spapr_max_cores
);
3807 return machine
->possible_cpus
;
3810 core_type
= spapr_get_cpu_core_type(machine
->cpu_type
);
3812 error_report("Unable to find sPAPR CPU Core definition");
3816 machine
->possible_cpus
= g_malloc0(sizeof(CPUArchIdList
) +
3817 sizeof(CPUArchId
) * spapr_max_cores
);
3818 machine
->possible_cpus
->len
= spapr_max_cores
;
3819 for (i
= 0; i
< machine
->possible_cpus
->len
; i
++) {
3820 int core_id
= i
* smp_threads
;
3822 machine
->possible_cpus
->cpus
[i
].type
= core_type
;
3823 machine
->possible_cpus
->cpus
[i
].vcpus_count
= smp_threads
;
3824 machine
->possible_cpus
->cpus
[i
].arch_id
= core_id
;
3825 machine
->possible_cpus
->cpus
[i
].props
.has_core_id
= true;
3826 machine
->possible_cpus
->cpus
[i
].props
.core_id
= core_id
;
3828 return machine
->possible_cpus
;
3831 static void spapr_phb_placement(sPAPRMachineState
*spapr
, uint32_t index
,
3832 uint64_t *buid
, hwaddr
*pio
,
3833 hwaddr
*mmio32
, hwaddr
*mmio64
,
3834 unsigned n_dma
, uint32_t *liobns
, Error
**errp
)
3837 * New-style PHB window placement.
3839 * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3840 * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3843 * Some guest kernels can't work with MMIO windows above 1<<46
3844 * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3846 * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3847 * PHB stacked together. (32TiB+2GiB)..(32TiB+64GiB) contains the
3848 * 2GiB 32-bit MMIO windows for each PHB. Then 33..64TiB has the
3849 * 1TiB 64-bit MMIO windows for each PHB.
3851 const uint64_t base_buid
= 0x800000020000000ULL
;
3854 /* Sanity check natural alignments */
3855 QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE
% SPAPR_PCI_MEM64_WIN_SIZE
) != 0);
3856 QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT
% SPAPR_PCI_MEM64_WIN_SIZE
) != 0);
3857 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE
% SPAPR_PCI_MEM32_WIN_SIZE
) != 0);
3858 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE
% SPAPR_PCI_IO_WIN_SIZE
) != 0);
3859 /* Sanity check bounds */
3860 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS
* SPAPR_PCI_IO_WIN_SIZE
) >
3861 SPAPR_PCI_MEM32_WIN_SIZE
);
3862 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS
* SPAPR_PCI_MEM32_WIN_SIZE
) >
3863 SPAPR_PCI_MEM64_WIN_SIZE
);
3865 if (index
>= SPAPR_MAX_PHBS
) {
3866 error_setg(errp
, "\"index\" for PAPR PHB is too large (max %llu)",
3867 SPAPR_MAX_PHBS
- 1);
3871 *buid
= base_buid
+ index
;
3872 for (i
= 0; i
< n_dma
; ++i
) {
3873 liobns
[i
] = SPAPR_PCI_LIOBN(index
, i
);
3876 *pio
= SPAPR_PCI_BASE
+ index
* SPAPR_PCI_IO_WIN_SIZE
;
3877 *mmio32
= SPAPR_PCI_BASE
+ (index
+ 1) * SPAPR_PCI_MEM32_WIN_SIZE
;
3878 *mmio64
= SPAPR_PCI_BASE
+ (index
+ 1) * SPAPR_PCI_MEM64_WIN_SIZE
;
3881 static ICSState
*spapr_ics_get(XICSFabric
*dev
, int irq
)
3883 sPAPRMachineState
*spapr
= SPAPR_MACHINE(dev
);
3885 return ics_valid_irq(spapr
->ics
, irq
) ? spapr
->ics
: NULL
;
3888 static void spapr_ics_resend(XICSFabric
*dev
)
3890 sPAPRMachineState
*spapr
= SPAPR_MACHINE(dev
);
3892 ics_resend(spapr
->ics
);
3895 static ICPState
*spapr_icp_get(XICSFabric
*xi
, int vcpu_id
)
3897 PowerPCCPU
*cpu
= spapr_find_cpu(vcpu_id
);
3899 return cpu
? cpu
->icp
: NULL
;
3902 static void spapr_pic_print_info(InterruptStatsProvider
*obj
,
3905 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
3907 spapr
->irq
->print_info(spapr
, mon
);
3910 int spapr_get_vcpu_id(PowerPCCPU
*cpu
)
3912 return cpu
->vcpu_id
;
3915 void spapr_set_vcpu_id(PowerPCCPU
*cpu
, int cpu_index
, Error
**errp
)
3917 sPAPRMachineState
*spapr
= SPAPR_MACHINE(qdev_get_machine());
3920 vcpu_id
= spapr_vcpu_id(spapr
, cpu_index
);
3922 if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id
)) {
3923 error_setg(errp
, "Can't create CPU with id %d in KVM", vcpu_id
);
3924 error_append_hint(errp
, "Adjust the number of cpus to %d "
3925 "or try to raise the number of threads per core\n",
3926 vcpu_id
* smp_threads
/ spapr
->vsmt
);
3930 cpu
->vcpu_id
= vcpu_id
;
3933 PowerPCCPU
*spapr_find_cpu(int vcpu_id
)
3938 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
3940 if (spapr_get_vcpu_id(cpu
) == vcpu_id
) {
3948 static void spapr_machine_class_init(ObjectClass
*oc
, void *data
)
3950 MachineClass
*mc
= MACHINE_CLASS(oc
);
3951 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(oc
);
3952 FWPathProviderClass
*fwc
= FW_PATH_PROVIDER_CLASS(oc
);
3953 NMIClass
*nc
= NMI_CLASS(oc
);
3954 HotplugHandlerClass
*hc
= HOTPLUG_HANDLER_CLASS(oc
);
3955 PPCVirtualHypervisorClass
*vhc
= PPC_VIRTUAL_HYPERVISOR_CLASS(oc
);
3956 XICSFabricClass
*xic
= XICS_FABRIC_CLASS(oc
);
3957 InterruptStatsProviderClass
*ispc
= INTERRUPT_STATS_PROVIDER_CLASS(oc
);
3959 mc
->desc
= "pSeries Logical Partition (PAPR compliant)";
3960 mc
->ignore_boot_device_suffixes
= true;
3963 * We set up the default / latest behaviour here. The class_init
3964 * functions for the specific versioned machine types can override
3965 * these details for backwards compatibility
3967 mc
->init
= spapr_machine_init
;
3968 mc
->reset
= spapr_machine_reset
;
3969 mc
->block_default_type
= IF_SCSI
;
3970 mc
->max_cpus
= 1024;
3971 mc
->no_parallel
= 1;
3972 mc
->default_boot_order
= "";
3973 mc
->default_ram_size
= 512 * MiB
;
3974 mc
->default_display
= "std";
3975 mc
->kvm_type
= spapr_kvm_type
;
3976 machine_class_allow_dynamic_sysbus_dev(mc
, TYPE_SPAPR_PCI_HOST_BRIDGE
);
3977 mc
->pci_allow_0_address
= true;
3978 assert(!mc
->get_hotplug_handler
);
3979 mc
->get_hotplug_handler
= spapr_get_hotplug_handler
;
3980 hc
->pre_plug
= spapr_machine_device_pre_plug
;
3981 hc
->plug
= spapr_machine_device_plug
;
3982 mc
->cpu_index_to_instance_props
= spapr_cpu_index_to_props
;
3983 mc
->get_default_cpu_node_id
= spapr_get_default_cpu_node_id
;
3984 mc
->possible_cpu_arch_ids
= spapr_possible_cpu_arch_ids
;
3985 hc
->unplug_request
= spapr_machine_device_unplug_request
;
3986 hc
->unplug
= spapr_machine_device_unplug
;
3988 smc
->dr_lmb_enabled
= true;
3989 smc
->update_dt_enabled
= true;
3990 mc
->default_cpu_type
= POWERPC_CPU_TYPE_NAME("power9_v2.0");
3991 mc
->has_hotpluggable_cpus
= true;
3992 smc
->resize_hpt_default
= SPAPR_RESIZE_HPT_ENABLED
;
3993 fwc
->get_dev_path
= spapr_get_fw_dev_path
;
3994 nc
->nmi_monitor_handler
= spapr_nmi
;
3995 smc
->phb_placement
= spapr_phb_placement
;
3996 vhc
->hypercall
= emulate_spapr_hypercall
;
3997 vhc
->hpt_mask
= spapr_hpt_mask
;
3998 vhc
->map_hptes
= spapr_map_hptes
;
3999 vhc
->unmap_hptes
= spapr_unmap_hptes
;
4000 vhc
->store_hpte
= spapr_store_hpte
;
4001 vhc
->get_patbe
= spapr_get_patbe
;
4002 vhc
->encode_hpt_for_kvm_pr
= spapr_encode_hpt_for_kvm_pr
;
4003 xic
->ics_get
= spapr_ics_get
;
4004 xic
->ics_resend
= spapr_ics_resend
;
4005 xic
->icp_get
= spapr_icp_get
;
4006 ispc
->print_info
= spapr_pic_print_info
;
4007 /* Force NUMA node memory size to be a multiple of
4008 * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4009 * in which LMBs are represented and hot-added
4011 mc
->numa_mem_align_shift
= 28;
4013 smc
->default_caps
.caps
[SPAPR_CAP_HTM
] = SPAPR_CAP_OFF
;
4014 smc
->default_caps
.caps
[SPAPR_CAP_VSX
] = SPAPR_CAP_ON
;
4015 smc
->default_caps
.caps
[SPAPR_CAP_DFP
] = SPAPR_CAP_ON
;
4016 smc
->default_caps
.caps
[SPAPR_CAP_CFPC
] = SPAPR_CAP_BROKEN
;
4017 smc
->default_caps
.caps
[SPAPR_CAP_SBBC
] = SPAPR_CAP_BROKEN
;
4018 smc
->default_caps
.caps
[SPAPR_CAP_IBS
] = SPAPR_CAP_BROKEN
;
4019 smc
->default_caps
.caps
[SPAPR_CAP_HPT_MAXPAGESIZE
] = 16; /* 64kiB */
4020 smc
->default_caps
.caps
[SPAPR_CAP_NESTED_KVM_HV
] = SPAPR_CAP_OFF
;
4021 spapr_caps_add_properties(smc
, &error_abort
);
4022 smc
->irq
= &spapr_irq_xics
;
4025 static const TypeInfo spapr_machine_info
= {
4026 .name
= TYPE_SPAPR_MACHINE
,
4027 .parent
= TYPE_MACHINE
,
4029 .instance_size
= sizeof(sPAPRMachineState
),
4030 .instance_init
= spapr_instance_init
,
4031 .instance_finalize
= spapr_machine_finalizefn
,
4032 .class_size
= sizeof(sPAPRMachineClass
),
4033 .class_init
= spapr_machine_class_init
,
4034 .interfaces
= (InterfaceInfo
[]) {
4035 { TYPE_FW_PATH_PROVIDER
},
4037 { TYPE_HOTPLUG_HANDLER
},
4038 { TYPE_PPC_VIRTUAL_HYPERVISOR
},
4039 { TYPE_XICS_FABRIC
},
4040 { TYPE_INTERRUPT_STATS_PROVIDER
},
4045 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest) \
4046 static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4049 MachineClass *mc = MACHINE_CLASS(oc); \
4050 spapr_machine_##suffix##_class_options(mc); \
4052 mc->alias = "pseries"; \
4053 mc->is_default = 1; \
4056 static const TypeInfo spapr_machine_##suffix##_info = { \
4057 .name = MACHINE_TYPE_NAME("pseries-" verstr), \
4058 .parent = TYPE_SPAPR_MACHINE, \
4059 .class_init = spapr_machine_##suffix##_class_init, \
4061 static void spapr_machine_register_##suffix(void) \
4063 type_register(&spapr_machine_##suffix##_info); \
4065 type_init(spapr_machine_register_##suffix)
4070 static void spapr_machine_4_0_class_options(MachineClass
*mc
)
4072 /* Defaults for the latest behaviour inherited from the base class */
4075 DEFINE_SPAPR_MACHINE(4_0
, "4.0", true);
4080 static void spapr_machine_3_1_class_options(MachineClass
*mc
)
4082 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4084 spapr_machine_4_0_class_options(mc
);
4085 compat_props_add(mc
->compat_props
, hw_compat_3_1
, hw_compat_3_1_len
);
4086 mc
->default_cpu_type
= POWERPC_CPU_TYPE_NAME("power8_v2.0");
4087 smc
->update_dt_enabled
= false;
4090 DEFINE_SPAPR_MACHINE(3_1
, "3.1", false);
4096 static void spapr_machine_3_0_class_options(MachineClass
*mc
)
4098 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4100 spapr_machine_3_1_class_options(mc
);
4101 compat_props_add(mc
->compat_props
, hw_compat_3_0
, hw_compat_3_0_len
);
4103 smc
->legacy_irq_allocation
= true;
4104 smc
->irq
= &spapr_irq_xics_legacy
;
4107 DEFINE_SPAPR_MACHINE(3_0
, "3.0", false);
4112 static void spapr_machine_2_12_class_options(MachineClass
*mc
)
4114 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4115 static GlobalProperty compat
[] = {
4117 .driver
= TYPE_POWERPC_CPU
,
4118 .property
= "pre-3.0-migration",
4122 .driver
= TYPE_SPAPR_CPU_CORE
,
4123 .property
= "pre-3.0-migration",
4128 spapr_machine_3_0_class_options(mc
);
4129 compat_props_add(mc
->compat_props
, hw_compat_2_12
, hw_compat_2_12_len
);
4130 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
4132 /* We depend on kvm_enabled() to choose a default value for the
4133 * hpt-max-page-size capability. Of course we can't do it here
4134 * because this is too early and the HW accelerator isn't initialzed
4135 * yet. Postpone this to machine init (see default_caps_with_cpu()).
4137 smc
->default_caps
.caps
[SPAPR_CAP_HPT_MAXPAGESIZE
] = 0;
4140 DEFINE_SPAPR_MACHINE(2_12
, "2.12", false);
4142 static void spapr_machine_2_12_sxxm_class_options(MachineClass
*mc
)
4144 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4146 spapr_machine_2_12_class_options(mc
);
4147 smc
->default_caps
.caps
[SPAPR_CAP_CFPC
] = SPAPR_CAP_WORKAROUND
;
4148 smc
->default_caps
.caps
[SPAPR_CAP_SBBC
] = SPAPR_CAP_WORKAROUND
;
4149 smc
->default_caps
.caps
[SPAPR_CAP_IBS
] = SPAPR_CAP_FIXED_CCD
;
4152 DEFINE_SPAPR_MACHINE(2_12_sxxm
, "2.12-sxxm", false);
4158 static void spapr_machine_2_11_class_options(MachineClass
*mc
)
4160 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4162 spapr_machine_2_12_class_options(mc
);
4163 smc
->default_caps
.caps
[SPAPR_CAP_HTM
] = SPAPR_CAP_ON
;
4164 compat_props_add(mc
->compat_props
, hw_compat_2_11
, hw_compat_2_11_len
);
4167 DEFINE_SPAPR_MACHINE(2_11
, "2.11", false);
4173 static void spapr_machine_2_10_class_options(MachineClass
*mc
)
4175 spapr_machine_2_11_class_options(mc
);
4176 compat_props_add(mc
->compat_props
, hw_compat_2_10
, hw_compat_2_10_len
);
4179 DEFINE_SPAPR_MACHINE(2_10
, "2.10", false);
4185 static void spapr_machine_2_9_class_options(MachineClass
*mc
)
4187 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4188 static GlobalProperty compat
[] = {
4190 .driver
= TYPE_POWERPC_CPU
,
4191 .property
= "pre-2.10-migration",
4196 spapr_machine_2_10_class_options(mc
);
4197 compat_props_add(mc
->compat_props
, hw_compat_2_9
, hw_compat_2_9_len
);
4198 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
4199 mc
->numa_auto_assign_ram
= numa_legacy_auto_assign_ram
;
4200 smc
->pre_2_10_has_unused_icps
= true;
4201 smc
->resize_hpt_default
= SPAPR_RESIZE_HPT_DISABLED
;
4204 DEFINE_SPAPR_MACHINE(2_9
, "2.9", false);
4210 static void spapr_machine_2_8_class_options(MachineClass
*mc
)
4212 static GlobalProperty compat
[] = {
4214 .driver
= TYPE_SPAPR_PCI_HOST_BRIDGE
,
4215 .property
= "pcie-extended-configuration-space",
4220 spapr_machine_2_9_class_options(mc
);
4221 compat_props_add(mc
->compat_props
, hw_compat_2_8
, hw_compat_2_8_len
);
4222 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
4223 mc
->numa_mem_align_shift
= 23;
4226 DEFINE_SPAPR_MACHINE(2_8
, "2.8", false);
4232 static void phb_placement_2_7(sPAPRMachineState
*spapr
, uint32_t index
,
4233 uint64_t *buid
, hwaddr
*pio
,
4234 hwaddr
*mmio32
, hwaddr
*mmio64
,
4235 unsigned n_dma
, uint32_t *liobns
, Error
**errp
)
4237 /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4238 const uint64_t base_buid
= 0x800000020000000ULL
;
4239 const hwaddr phb_spacing
= 0x1000000000ULL
; /* 64 GiB */
4240 const hwaddr mmio_offset
= 0xa0000000; /* 2 GiB + 512 MiB */
4241 const hwaddr pio_offset
= 0x80000000; /* 2 GiB */
4242 const uint32_t max_index
= 255;
4243 const hwaddr phb0_alignment
= 0x10000000000ULL
; /* 1 TiB */
4245 uint64_t ram_top
= MACHINE(spapr
)->ram_size
;
4246 hwaddr phb0_base
, phb_base
;
4249 /* Do we have device memory? */
4250 if (MACHINE(spapr
)->maxram_size
> ram_top
) {
4251 /* Can't just use maxram_size, because there may be an
4252 * alignment gap between normal and device memory regions
4254 ram_top
= MACHINE(spapr
)->device_memory
->base
+
4255 memory_region_size(&MACHINE(spapr
)->device_memory
->mr
);
4258 phb0_base
= QEMU_ALIGN_UP(ram_top
, phb0_alignment
);
4260 if (index
> max_index
) {
4261 error_setg(errp
, "\"index\" for PAPR PHB is too large (max %u)",
4266 *buid
= base_buid
+ index
;
4267 for (i
= 0; i
< n_dma
; ++i
) {
4268 liobns
[i
] = SPAPR_PCI_LIOBN(index
, i
);
4271 phb_base
= phb0_base
+ index
* phb_spacing
;
4272 *pio
= phb_base
+ pio_offset
;
4273 *mmio32
= phb_base
+ mmio_offset
;
4275 * We don't set the 64-bit MMIO window, relying on the PHB's
4276 * fallback behaviour of automatically splitting a large "32-bit"
4277 * window into contiguous 32-bit and 64-bit windows
4281 static void spapr_machine_2_7_class_options(MachineClass
*mc
)
4283 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4284 static GlobalProperty compat
[] = {
4286 .driver
= TYPE_SPAPR_PCI_HOST_BRIDGE
,
4287 .property
= "mem_win_size",
4288 .value
= stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE
),
4291 .driver
= TYPE_SPAPR_PCI_HOST_BRIDGE
,
4292 .property
= "mem64_win_size",
4296 .driver
= TYPE_POWERPC_CPU
,
4297 .property
= "pre-2.8-migration",
4301 .driver
= TYPE_SPAPR_PCI_HOST_BRIDGE
,
4302 .property
= "pre-2.8-migration",
4307 spapr_machine_2_8_class_options(mc
);
4308 mc
->default_cpu_type
= POWERPC_CPU_TYPE_NAME("power7_v2.3");
4309 mc
->default_machine_opts
= "modern-hotplug-events=off";
4310 compat_props_add(mc
->compat_props
, hw_compat_2_7
, hw_compat_2_7_len
);
4311 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
4312 smc
->phb_placement
= phb_placement_2_7
;
4315 DEFINE_SPAPR_MACHINE(2_7
, "2.7", false);
4321 static void spapr_machine_2_6_class_options(MachineClass
*mc
)
4323 static GlobalProperty compat
[] = {
4325 .driver
= TYPE_SPAPR_PCI_HOST_BRIDGE
,
4327 .value
= stringify(off
),
4331 spapr_machine_2_7_class_options(mc
);
4332 mc
->has_hotpluggable_cpus
= false;
4333 compat_props_add(mc
->compat_props
, hw_compat_2_6
, hw_compat_2_6_len
);
4334 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
4337 DEFINE_SPAPR_MACHINE(2_6
, "2.6", false);
4343 static void spapr_machine_2_5_class_options(MachineClass
*mc
)
4345 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4346 static GlobalProperty compat
[] = {
4348 .driver
= "spapr-vlan",
4349 .property
= "use-rx-buffer-pools",
4354 spapr_machine_2_6_class_options(mc
);
4355 smc
->use_ohci_by_default
= true;
4356 compat_props_add(mc
->compat_props
, hw_compat_2_5
, hw_compat_2_5_len
);
4357 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
4360 DEFINE_SPAPR_MACHINE(2_5
, "2.5", false);
4366 static void spapr_machine_2_4_class_options(MachineClass
*mc
)
4368 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
4370 spapr_machine_2_5_class_options(mc
);
4371 smc
->dr_lmb_enabled
= false;
4372 compat_props_add(mc
->compat_props
, hw_compat_2_4
, hw_compat_2_4_len
);
4375 DEFINE_SPAPR_MACHINE(2_4
, "2.4", false);
4381 static void spapr_machine_2_3_class_options(MachineClass
*mc
)
4383 static GlobalProperty compat
[] = {
4385 .driver
= "spapr-pci-host-bridge",
4386 .property
= "dynamic-reconfiguration",
4390 spapr_machine_2_4_class_options(mc
);
4391 compat_props_add(mc
->compat_props
, hw_compat_2_3
, hw_compat_2_3_len
);
4392 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
4394 DEFINE_SPAPR_MACHINE(2_3
, "2.3", false);
4400 static void spapr_machine_2_2_class_options(MachineClass
*mc
)
4402 static GlobalProperty compat
[] = {
4404 .driver
= TYPE_SPAPR_PCI_HOST_BRIDGE
,
4405 .property
= "mem_win_size",
4406 .value
= "0x20000000",
4410 spapr_machine_2_3_class_options(mc
);
4411 compat_props_add(mc
->compat_props
, hw_compat_2_2
, hw_compat_2_2_len
);
4412 compat_props_add(mc
->compat_props
, compat
, G_N_ELEMENTS(compat
));
4413 mc
->default_machine_opts
= "modern-hotplug-events=off,suppress-vmdesc=on";
4415 DEFINE_SPAPR_MACHINE(2_2
, "2.2", false);
4421 static void spapr_machine_2_1_class_options(MachineClass
*mc
)
4423 spapr_machine_2_2_class_options(mc
);
4424 compat_props_add(mc
->compat_props
, hw_compat_2_1
, hw_compat_2_1_len
);
4426 DEFINE_SPAPR_MACHINE(2_1
, "2.1", false);
4428 static void spapr_machine_register_types(void)
4430 type_register_static(&spapr_machine_info
);
4433 type_init(spapr_machine_register_types
)