2 * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
4 * Copyright (c) 2004-2007 Fabrice Bellard
5 * Copyright (c) 2007 Jocelyn Mayer
6 * Copyright (c) 2010 David Gibson, IBM Corporation.
8 * Permission is hereby granted, free of charge, to any person obtaining a copy
9 * of this software and associated documentation files (the "Software"), to deal
10 * in the Software without restriction, including without limitation the rights
11 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12 * copies of the Software, and to permit persons to whom the Software is
13 * furnished to do so, subject to the following conditions:
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
27 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "sysemu/sysemu.h"
30 #include "sysemu/numa.h"
33 #include "hw/fw-path-provider.h"
36 #include "sysemu/device_tree.h"
37 #include "sysemu/block-backend.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
41 #include "migration/misc.h"
42 #include "migration/global_state.h"
43 #include "migration/register.h"
44 #include "mmu-hash64.h"
45 #include "mmu-book3s-v3.h"
48 #include "hw/boards.h"
49 #include "hw/ppc/ppc.h"
50 #include "hw/loader.h"
52 #include "hw/ppc/fdt.h"
53 #include "hw/ppc/spapr.h"
54 #include "hw/ppc/spapr_vio.h"
55 #include "hw/pci-host/spapr.h"
56 #include "hw/ppc/xics.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"
66 #include "qemu/config-file.h"
67 #include "qemu/error-report.h"
70 #include "hw/intc/intc.h"
72 #include "hw/compat.h"
73 #include "qemu/cutils.h"
74 #include "hw/ppc/spapr_cpu_core.h"
75 #include "qmp-commands.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 #define HTAB_SIZE(spapr) (1ULL << ((spapr)->htab_shift))
103 static ICSState
*spapr_ics_create(sPAPRMachineState
*spapr
,
104 const char *type_ics
,
105 int nr_irqs
, Error
**errp
)
107 Error
*local_err
= NULL
;
110 obj
= object_new(type_ics
);
111 object_property_add_child(OBJECT(spapr
), "ics", obj
, &error_abort
);
112 object_property_add_const_link(obj
, ICS_PROP_XICS
, OBJECT(spapr
),
114 object_property_set_int(obj
, nr_irqs
, "nr-irqs", &local_err
);
118 object_property_set_bool(obj
, true, "realized", &local_err
);
123 return ICS_SIMPLE(obj
);
126 error_propagate(errp
, local_err
);
130 static void xics_system_init(MachineState
*machine
, int nr_irqs
, Error
**errp
)
132 sPAPRMachineState
*spapr
= SPAPR_MACHINE(machine
);
135 if (machine_kernel_irqchip_allowed(machine
) &&
136 !xics_kvm_init(spapr
, errp
)) {
137 spapr
->icp_type
= TYPE_KVM_ICP
;
138 spapr
->ics
= spapr_ics_create(spapr
, TYPE_ICS_KVM
, nr_irqs
, errp
);
140 if (machine_kernel_irqchip_required(machine
) && !spapr
->ics
) {
141 error_prepend(errp
, "kernel_irqchip requested but unavailable: ");
147 xics_spapr_init(spapr
);
148 spapr
->icp_type
= TYPE_ICP
;
149 spapr
->ics
= spapr_ics_create(spapr
, TYPE_ICS_SIMPLE
, nr_irqs
, errp
);
156 static int spapr_fixup_cpu_smt_dt(void *fdt
, int offset
, PowerPCCPU
*cpu
,
160 uint32_t servers_prop
[smt_threads
];
161 uint32_t gservers_prop
[smt_threads
* 2];
162 int index
= ppc_get_vcpu_dt_id(cpu
);
164 if (cpu
->compat_pvr
) {
165 ret
= fdt_setprop_cell(fdt
, offset
, "cpu-version", cpu
->compat_pvr
);
171 /* Build interrupt servers and gservers properties */
172 for (i
= 0; i
< smt_threads
; i
++) {
173 servers_prop
[i
] = cpu_to_be32(index
+ i
);
174 /* Hack, direct the group queues back to cpu 0 */
175 gservers_prop
[i
*2] = cpu_to_be32(index
+ i
);
176 gservers_prop
[i
*2 + 1] = 0;
178 ret
= fdt_setprop(fdt
, offset
, "ibm,ppc-interrupt-server#s",
179 servers_prop
, sizeof(servers_prop
));
183 ret
= fdt_setprop(fdt
, offset
, "ibm,ppc-interrupt-gserver#s",
184 gservers_prop
, sizeof(gservers_prop
));
189 static int spapr_fixup_cpu_numa_dt(void *fdt
, int offset
, PowerPCCPU
*cpu
)
191 int index
= ppc_get_vcpu_dt_id(cpu
);
192 uint32_t associativity
[] = {cpu_to_be32(0x5),
196 cpu_to_be32(cpu
->node_id
),
199 /* Advertise NUMA via ibm,associativity */
200 return fdt_setprop(fdt
, offset
, "ibm,associativity", associativity
,
201 sizeof(associativity
));
204 /* Populate the "ibm,pa-features" property */
205 static void spapr_populate_pa_features(CPUPPCState
*env
, void *fdt
, int offset
,
208 uint8_t pa_features_206
[] = { 6, 0,
209 0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
210 uint8_t pa_features_207
[] = { 24, 0,
211 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
212 0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
213 0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
214 0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
215 uint8_t pa_features_300
[] = { 66, 0,
216 /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
217 /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
218 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
220 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
222 0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
223 /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
224 0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
225 /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
226 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
227 /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
228 0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
229 /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
230 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
231 /* 42: PM, 44: PC RA, 46: SC vec'd */
232 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
233 /* 48: SIMD, 50: QP BFP, 52: String */
234 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
235 /* 54: DecFP, 56: DecI, 58: SHA */
236 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
237 /* 60: NM atomic, 62: RNG */
238 0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
240 uint8_t *pa_features
;
243 switch (POWERPC_MMU_VER(env
->mmu_model
)) {
244 case POWERPC_MMU_VER_2_06
:
245 pa_features
= pa_features_206
;
246 pa_size
= sizeof(pa_features_206
);
248 case POWERPC_MMU_VER_2_07
:
249 pa_features
= pa_features_207
;
250 pa_size
= sizeof(pa_features_207
);
252 case POWERPC_MMU_VER_3_00
:
253 pa_features
= pa_features_300
;
254 pa_size
= sizeof(pa_features_300
);
260 if (env
->ci_large_pages
) {
262 * Note: we keep CI large pages off by default because a 64K capable
263 * guest provisioned with large pages might otherwise try to map a qemu
264 * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
265 * even if that qemu runs on a 4k host.
266 * We dd this bit back here if we are confident this is not an issue
268 pa_features
[3] |= 0x20;
270 if (kvmppc_has_cap_htm() && pa_size
> 24) {
271 pa_features
[24] |= 0x80; /* Transactional memory support */
273 if (legacy_guest
&& pa_size
> 40) {
274 /* Workaround for broken kernels that attempt (guest) radix
275 * mode when they can't handle it, if they see the radix bit set
276 * in pa-features. So hide it from them. */
277 pa_features
[40 + 2] &= ~0x80; /* Radix MMU */
280 _FDT((fdt_setprop(fdt
, offset
, "ibm,pa-features", pa_features
, pa_size
)));
283 static int spapr_fixup_cpu_dt(void *fdt
, sPAPRMachineState
*spapr
)
285 int ret
= 0, offset
, cpus_offset
;
288 int smt
= kvmppc_smt_threads();
289 uint32_t pft_size_prop
[] = {0, cpu_to_be32(spapr
->htab_shift
)};
292 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
293 CPUPPCState
*env
= &cpu
->env
;
294 DeviceClass
*dc
= DEVICE_GET_CLASS(cs
);
295 int index
= ppc_get_vcpu_dt_id(cpu
);
296 int compat_smt
= MIN(smp_threads
, ppc_compat_max_threads(cpu
));
298 if ((index
% smt
) != 0) {
302 snprintf(cpu_model
, 32, "%s@%x", dc
->fw_name
, index
);
304 cpus_offset
= fdt_path_offset(fdt
, "/cpus");
305 if (cpus_offset
< 0) {
306 cpus_offset
= fdt_add_subnode(fdt
, fdt_path_offset(fdt
, "/"),
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(env
, fdt
, offset
,
339 spapr
->cas_legacy_guest_workaround
);
344 static hwaddr
spapr_node0_size(void)
346 MachineState
*machine
= MACHINE(qdev_get_machine());
350 for (i
= 0; i
< nb_numa_nodes
; ++i
) {
351 if (numa_info
[i
].node_mem
) {
352 return MIN(pow2floor(numa_info
[i
].node_mem
),
357 return machine
->ram_size
;
360 static void add_str(GString
*s
, const gchar
*s1
)
362 g_string_append_len(s
, s1
, strlen(s1
) + 1);
365 static int spapr_populate_memory_node(void *fdt
, int nodeid
, hwaddr start
,
368 uint32_t associativity
[] = {
369 cpu_to_be32(0x4), /* length */
370 cpu_to_be32(0x0), cpu_to_be32(0x0),
371 cpu_to_be32(0x0), cpu_to_be32(nodeid
)
374 uint64_t mem_reg_property
[2];
377 mem_reg_property
[0] = cpu_to_be64(start
);
378 mem_reg_property
[1] = cpu_to_be64(size
);
380 sprintf(mem_name
, "memory@" TARGET_FMT_lx
, start
);
381 off
= fdt_add_subnode(fdt
, 0, mem_name
);
383 _FDT((fdt_setprop_string(fdt
, off
, "device_type", "memory")));
384 _FDT((fdt_setprop(fdt
, off
, "reg", mem_reg_property
,
385 sizeof(mem_reg_property
))));
386 _FDT((fdt_setprop(fdt
, off
, "ibm,associativity", associativity
,
387 sizeof(associativity
))));
391 static int spapr_populate_memory(sPAPRMachineState
*spapr
, void *fdt
)
393 MachineState
*machine
= MACHINE(spapr
);
394 hwaddr mem_start
, node_size
;
395 int i
, nb_nodes
= nb_numa_nodes
;
396 NodeInfo
*nodes
= numa_info
;
399 /* No NUMA nodes, assume there is just one node with whole RAM */
400 if (!nb_numa_nodes
) {
402 ramnode
.node_mem
= machine
->ram_size
;
406 for (i
= 0, mem_start
= 0; i
< nb_nodes
; ++i
) {
407 if (!nodes
[i
].node_mem
) {
410 if (mem_start
>= machine
->ram_size
) {
413 node_size
= nodes
[i
].node_mem
;
414 if (node_size
> machine
->ram_size
- mem_start
) {
415 node_size
= machine
->ram_size
- mem_start
;
419 /* ppc_spapr_init() checks for rma_size <= node0_size already */
420 spapr_populate_memory_node(fdt
, i
, 0, spapr
->rma_size
);
421 mem_start
+= spapr
->rma_size
;
422 node_size
-= spapr
->rma_size
;
424 for ( ; node_size
; ) {
425 hwaddr sizetmp
= pow2floor(node_size
);
427 /* mem_start != 0 here */
428 if (ctzl(mem_start
) < ctzl(sizetmp
)) {
429 sizetmp
= 1ULL << ctzl(mem_start
);
432 spapr_populate_memory_node(fdt
, i
, mem_start
, sizetmp
);
433 node_size
-= sizetmp
;
434 mem_start
+= sizetmp
;
441 static void spapr_populate_cpu_dt(CPUState
*cs
, void *fdt
, int offset
,
442 sPAPRMachineState
*spapr
)
444 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
445 CPUPPCState
*env
= &cpu
->env
;
446 PowerPCCPUClass
*pcc
= POWERPC_CPU_GET_CLASS(cs
);
447 int index
= ppc_get_vcpu_dt_id(cpu
);
448 uint32_t segs
[] = {cpu_to_be32(28), cpu_to_be32(40),
449 0xffffffff, 0xffffffff};
450 uint32_t tbfreq
= kvm_enabled() ? kvmppc_get_tbfreq()
451 : SPAPR_TIMEBASE_FREQ
;
452 uint32_t cpufreq
= kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
453 uint32_t page_sizes_prop
[64];
454 size_t page_sizes_prop_size
;
455 uint32_t vcpus_per_socket
= smp_threads
* smp_cores
;
456 uint32_t pft_size_prop
[] = {0, cpu_to_be32(spapr
->htab_shift
)};
457 int compat_smt
= MIN(smp_threads
, ppc_compat_max_threads(cpu
));
458 sPAPRDRConnector
*drc
;
460 uint32_t radix_AP_encodings
[PPC_PAGE_SIZES_MAX_SZ
];
463 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_CPU
, index
);
465 drc_index
= spapr_drc_index(drc
);
466 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,my-drc-index", drc_index
)));
469 _FDT((fdt_setprop_cell(fdt
, offset
, "reg", index
)));
470 _FDT((fdt_setprop_string(fdt
, offset
, "device_type", "cpu")));
472 _FDT((fdt_setprop_cell(fdt
, offset
, "cpu-version", env
->spr
[SPR_PVR
])));
473 _FDT((fdt_setprop_cell(fdt
, offset
, "d-cache-block-size",
474 env
->dcache_line_size
)));
475 _FDT((fdt_setprop_cell(fdt
, offset
, "d-cache-line-size",
476 env
->dcache_line_size
)));
477 _FDT((fdt_setprop_cell(fdt
, offset
, "i-cache-block-size",
478 env
->icache_line_size
)));
479 _FDT((fdt_setprop_cell(fdt
, offset
, "i-cache-line-size",
480 env
->icache_line_size
)));
482 if (pcc
->l1_dcache_size
) {
483 _FDT((fdt_setprop_cell(fdt
, offset
, "d-cache-size",
484 pcc
->l1_dcache_size
)));
486 error_report("Warning: Unknown L1 dcache size for cpu");
488 if (pcc
->l1_icache_size
) {
489 _FDT((fdt_setprop_cell(fdt
, offset
, "i-cache-size",
490 pcc
->l1_icache_size
)));
492 error_report("Warning: Unknown L1 icache size for cpu");
495 _FDT((fdt_setprop_cell(fdt
, offset
, "timebase-frequency", tbfreq
)));
496 _FDT((fdt_setprop_cell(fdt
, offset
, "clock-frequency", cpufreq
)));
497 _FDT((fdt_setprop_cell(fdt
, offset
, "slb-size", env
->slb_nr
)));
498 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,slb-size", env
->slb_nr
)));
499 _FDT((fdt_setprop_string(fdt
, offset
, "status", "okay")));
500 _FDT((fdt_setprop(fdt
, offset
, "64-bit", NULL
, 0)));
502 if (env
->spr_cb
[SPR_PURR
].oea_read
) {
503 _FDT((fdt_setprop(fdt
, offset
, "ibm,purr", NULL
, 0)));
506 if (env
->mmu_model
& POWERPC_MMU_1TSEG
) {
507 _FDT((fdt_setprop(fdt
, offset
, "ibm,processor-segment-sizes",
508 segs
, sizeof(segs
))));
511 /* Advertise VMX/VSX (vector extensions) if available
512 * 0 / no property == no vector extensions
513 * 1 == VMX / Altivec available
514 * 2 == VSX available */
515 if (env
->insns_flags
& PPC_ALTIVEC
) {
516 uint32_t vmx
= (env
->insns_flags2
& PPC2_VSX
) ? 2 : 1;
518 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,vmx", vmx
)));
521 /* Advertise DFP (Decimal Floating Point) if available
522 * 0 / no property == no DFP
523 * 1 == DFP available */
524 if (env
->insns_flags2
& PPC2_DFP
) {
525 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,dfp", 1)));
528 page_sizes_prop_size
= ppc_create_page_sizes_prop(env
, page_sizes_prop
,
529 sizeof(page_sizes_prop
));
530 if (page_sizes_prop_size
) {
531 _FDT((fdt_setprop(fdt
, offset
, "ibm,segment-page-sizes",
532 page_sizes_prop
, page_sizes_prop_size
)));
535 spapr_populate_pa_features(env
, fdt
, offset
, false);
537 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,chip-id",
538 cs
->cpu_index
/ vcpus_per_socket
)));
540 _FDT((fdt_setprop(fdt
, offset
, "ibm,pft-size",
541 pft_size_prop
, sizeof(pft_size_prop
))));
543 if (nb_numa_nodes
> 1) {
544 _FDT(spapr_fixup_cpu_numa_dt(fdt
, offset
, cpu
));
547 _FDT(spapr_fixup_cpu_smt_dt(fdt
, offset
, cpu
, compat_smt
));
549 if (pcc
->radix_page_info
) {
550 for (i
= 0; i
< pcc
->radix_page_info
->count
; i
++) {
551 radix_AP_encodings
[i
] =
552 cpu_to_be32(pcc
->radix_page_info
->entries
[i
]);
554 _FDT((fdt_setprop(fdt
, offset
, "ibm,processor-radix-AP-encodings",
556 pcc
->radix_page_info
->count
*
557 sizeof(radix_AP_encodings
[0]))));
561 static void spapr_populate_cpus_dt_node(void *fdt
, sPAPRMachineState
*spapr
)
566 int smt
= kvmppc_smt_threads();
568 cpus_offset
= fdt_add_subnode(fdt
, 0, "cpus");
570 _FDT((fdt_setprop_cell(fdt
, cpus_offset
, "#address-cells", 0x1)));
571 _FDT((fdt_setprop_cell(fdt
, cpus_offset
, "#size-cells", 0x0)));
574 * We walk the CPUs in reverse order to ensure that CPU DT nodes
575 * created by fdt_add_subnode() end up in the right order in FDT
576 * for the guest kernel the enumerate the CPUs correctly.
578 CPU_FOREACH_REVERSE(cs
) {
579 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
580 int index
= ppc_get_vcpu_dt_id(cpu
);
581 DeviceClass
*dc
= DEVICE_GET_CLASS(cs
);
584 if ((index
% smt
) != 0) {
588 nodename
= g_strdup_printf("%s@%x", dc
->fw_name
, index
);
589 offset
= fdt_add_subnode(fdt
, cpus_offset
, nodename
);
592 spapr_populate_cpu_dt(cs
, fdt
, offset
, spapr
);
598 * Adds ibm,dynamic-reconfiguration-memory node.
599 * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
600 * of this device tree node.
602 static int spapr_populate_drconf_memory(sPAPRMachineState
*spapr
, void *fdt
)
604 MachineState
*machine
= MACHINE(spapr
);
606 uint64_t lmb_size
= SPAPR_MEMORY_BLOCK_SIZE
;
607 uint32_t prop_lmb_size
[] = {0, cpu_to_be32(lmb_size
)};
608 uint32_t hotplug_lmb_start
= spapr
->hotplug_memory
.base
/ lmb_size
;
609 uint32_t nr_lmbs
= (spapr
->hotplug_memory
.base
+
610 memory_region_size(&spapr
->hotplug_memory
.mr
)) /
612 uint32_t *int_buf
, *cur_index
, buf_len
;
613 int nr_nodes
= nb_numa_nodes
? nb_numa_nodes
: 1;
616 * Don't create the node if there is no hotpluggable memory
618 if (machine
->ram_size
== machine
->maxram_size
) {
623 * Allocate enough buffer size to fit in ibm,dynamic-memory
624 * or ibm,associativity-lookup-arrays
626 buf_len
= MAX(nr_lmbs
* SPAPR_DR_LMB_LIST_ENTRY_SIZE
+ 1, nr_nodes
* 4 + 2)
628 cur_index
= int_buf
= g_malloc0(buf_len
);
630 offset
= fdt_add_subnode(fdt
, 0, "ibm,dynamic-reconfiguration-memory");
632 ret
= fdt_setprop(fdt
, offset
, "ibm,lmb-size", prop_lmb_size
,
633 sizeof(prop_lmb_size
));
638 ret
= fdt_setprop_cell(fdt
, offset
, "ibm,memory-flags-mask", 0xff);
643 ret
= fdt_setprop_cell(fdt
, offset
, "ibm,memory-preservation-time", 0x0);
648 /* ibm,dynamic-memory */
649 int_buf
[0] = cpu_to_be32(nr_lmbs
);
651 for (i
= 0; i
< nr_lmbs
; i
++) {
652 uint64_t addr
= i
* lmb_size
;
653 uint32_t *dynamic_memory
= cur_index
;
655 if (i
>= hotplug_lmb_start
) {
656 sPAPRDRConnector
*drc
;
658 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
, i
);
661 dynamic_memory
[0] = cpu_to_be32(addr
>> 32);
662 dynamic_memory
[1] = cpu_to_be32(addr
& 0xffffffff);
663 dynamic_memory
[2] = cpu_to_be32(spapr_drc_index(drc
));
664 dynamic_memory
[3] = cpu_to_be32(0); /* reserved */
665 dynamic_memory
[4] = cpu_to_be32(numa_get_node(addr
, NULL
));
666 if (memory_region_present(get_system_memory(), addr
)) {
667 dynamic_memory
[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED
);
669 dynamic_memory
[5] = cpu_to_be32(0);
673 * LMB information for RMA, boot time RAM and gap b/n RAM and
674 * hotplug memory region -- all these are marked as reserved
675 * and as having no valid DRC.
677 dynamic_memory
[0] = cpu_to_be32(addr
>> 32);
678 dynamic_memory
[1] = cpu_to_be32(addr
& 0xffffffff);
679 dynamic_memory
[2] = cpu_to_be32(0);
680 dynamic_memory
[3] = cpu_to_be32(0); /* reserved */
681 dynamic_memory
[4] = cpu_to_be32(-1);
682 dynamic_memory
[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED
|
683 SPAPR_LMB_FLAGS_DRC_INVALID
);
686 cur_index
+= SPAPR_DR_LMB_LIST_ENTRY_SIZE
;
688 ret
= fdt_setprop(fdt
, offset
, "ibm,dynamic-memory", int_buf
, buf_len
);
693 /* ibm,associativity-lookup-arrays */
695 int_buf
[0] = cpu_to_be32(nr_nodes
);
696 int_buf
[1] = cpu_to_be32(4); /* Number of entries per associativity list */
698 for (i
= 0; i
< nr_nodes
; i
++) {
699 uint32_t associativity
[] = {
705 memcpy(cur_index
, associativity
, sizeof(associativity
));
708 ret
= fdt_setprop(fdt
, offset
, "ibm,associativity-lookup-arrays", int_buf
,
709 (cur_index
- int_buf
) * sizeof(uint32_t));
715 static int spapr_dt_cas_updates(sPAPRMachineState
*spapr
, void *fdt
,
716 sPAPROptionVector
*ov5_updates
)
718 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(spapr
);
721 /* Generate ibm,dynamic-reconfiguration-memory node if required */
722 if (spapr_ovec_test(ov5_updates
, OV5_DRCONF_MEMORY
)) {
723 g_assert(smc
->dr_lmb_enabled
);
724 ret
= spapr_populate_drconf_memory(spapr
, fdt
);
730 offset
= fdt_path_offset(fdt
, "/chosen");
732 offset
= fdt_add_subnode(fdt
, 0, "chosen");
737 ret
= spapr_ovec_populate_dt(fdt
, offset
, spapr
->ov5_cas
,
738 "ibm,architecture-vec-5");
744 int spapr_h_cas_compose_response(sPAPRMachineState
*spapr
,
745 target_ulong addr
, target_ulong size
,
746 sPAPROptionVector
*ov5_updates
)
748 void *fdt
, *fdt_skel
;
749 sPAPRDeviceTreeUpdateHeader hdr
= { .version_id
= 1 };
753 /* Create sceleton */
754 fdt_skel
= g_malloc0(size
);
755 _FDT((fdt_create(fdt_skel
, size
)));
756 _FDT((fdt_begin_node(fdt_skel
, "")));
757 _FDT((fdt_end_node(fdt_skel
)));
758 _FDT((fdt_finish(fdt_skel
)));
759 fdt
= g_malloc0(size
);
760 _FDT((fdt_open_into(fdt_skel
, fdt
, size
)));
763 /* Fixup cpu nodes */
764 _FDT((spapr_fixup_cpu_dt(fdt
, spapr
)));
766 if (spapr_dt_cas_updates(spapr
, fdt
, ov5_updates
)) {
770 /* Pack resulting tree */
771 _FDT((fdt_pack(fdt
)));
773 if (fdt_totalsize(fdt
) + sizeof(hdr
) > size
) {
774 trace_spapr_cas_failed(size
);
778 cpu_physical_memory_write(addr
, &hdr
, sizeof(hdr
));
779 cpu_physical_memory_write(addr
+ sizeof(hdr
), fdt
, fdt_totalsize(fdt
));
780 trace_spapr_cas_continue(fdt_totalsize(fdt
) + sizeof(hdr
));
786 static void spapr_dt_rtas(sPAPRMachineState
*spapr
, void *fdt
)
789 GString
*hypertas
= g_string_sized_new(256);
790 GString
*qemu_hypertas
= g_string_sized_new(256);
791 uint32_t refpoints
[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
792 uint64_t max_hotplug_addr
= spapr
->hotplug_memory
.base
+
793 memory_region_size(&spapr
->hotplug_memory
.mr
);
794 uint32_t lrdr_capacity
[] = {
795 cpu_to_be32(max_hotplug_addr
>> 32),
796 cpu_to_be32(max_hotplug_addr
& 0xffffffff),
797 0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE
),
798 cpu_to_be32(max_cpus
/ smp_threads
),
801 _FDT(rtas
= fdt_add_subnode(fdt
, 0, "rtas"));
804 add_str(hypertas
, "hcall-pft");
805 add_str(hypertas
, "hcall-term");
806 add_str(hypertas
, "hcall-dabr");
807 add_str(hypertas
, "hcall-interrupt");
808 add_str(hypertas
, "hcall-tce");
809 add_str(hypertas
, "hcall-vio");
810 add_str(hypertas
, "hcall-splpar");
811 add_str(hypertas
, "hcall-bulk");
812 add_str(hypertas
, "hcall-set-mode");
813 add_str(hypertas
, "hcall-sprg0");
814 add_str(hypertas
, "hcall-copy");
815 add_str(hypertas
, "hcall-debug");
816 add_str(qemu_hypertas
, "hcall-memop1");
818 if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
819 add_str(hypertas
, "hcall-multi-tce");
821 _FDT(fdt_setprop(fdt
, rtas
, "ibm,hypertas-functions",
822 hypertas
->str
, hypertas
->len
));
823 g_string_free(hypertas
, TRUE
);
824 _FDT(fdt_setprop(fdt
, rtas
, "qemu,hypertas-functions",
825 qemu_hypertas
->str
, qemu_hypertas
->len
));
826 g_string_free(qemu_hypertas
, TRUE
);
828 _FDT(fdt_setprop(fdt
, rtas
, "ibm,associativity-reference-points",
829 refpoints
, sizeof(refpoints
)));
831 _FDT(fdt_setprop_cell(fdt
, rtas
, "rtas-error-log-max",
832 RTAS_ERROR_LOG_MAX
));
833 _FDT(fdt_setprop_cell(fdt
, rtas
, "rtas-event-scan-rate",
834 RTAS_EVENT_SCAN_RATE
));
837 _FDT(fdt_setprop(fdt
, rtas
, "ibm,change-msix-capable", NULL
, 0));
841 * According to PAPR, rtas ibm,os-term does not guarantee a return
842 * back to the guest cpu.
844 * While an additional ibm,extended-os-term property indicates
845 * that rtas call return will always occur. Set this property.
847 _FDT(fdt_setprop(fdt
, rtas
, "ibm,extended-os-term", NULL
, 0));
849 _FDT(fdt_setprop(fdt
, rtas
, "ibm,lrdr-capacity",
850 lrdr_capacity
, sizeof(lrdr_capacity
)));
852 spapr_dt_rtas_tokens(fdt
, rtas
);
855 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
856 * that the guest may request and thus the valid values for bytes 24..26 of
857 * option vector 5: */
858 static void spapr_dt_ov5_platform_support(void *fdt
, int chosen
)
860 PowerPCCPU
*first_ppc_cpu
= POWERPC_CPU(first_cpu
);
863 24, 0x00, /* Hash/Radix, filled in below. */
864 25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
865 26, 0x40, /* Radix options: GTSE == yes. */
869 if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
870 val
[1] = 0x80; /* OV5_MMU_BOTH */
871 } else if (kvmppc_has_cap_mmu_radix()) {
872 val
[1] = 0x40; /* OV5_MMU_RADIX_300 */
874 val
[1] = 0x00; /* Hash */
877 if (first_ppc_cpu
->env
.mmu_model
& POWERPC_MMU_V3
) {
878 /* V3 MMU supports both hash and radix (with dynamic switching) */
881 /* Otherwise we can only do hash */
885 _FDT(fdt_setprop(fdt
, chosen
, "ibm,arch-vec-5-platform-support",
889 static void spapr_dt_chosen(sPAPRMachineState
*spapr
, void *fdt
)
891 MachineState
*machine
= MACHINE(spapr
);
893 const char *boot_device
= machine
->boot_order
;
894 char *stdout_path
= spapr_vio_stdout_path(spapr
->vio_bus
);
896 char *bootlist
= get_boot_devices_list(&cb
, true);
898 _FDT(chosen
= fdt_add_subnode(fdt
, 0, "chosen"));
900 _FDT(fdt_setprop_string(fdt
, chosen
, "bootargs", machine
->kernel_cmdline
));
901 _FDT(fdt_setprop_cell(fdt
, chosen
, "linux,initrd-start",
902 spapr
->initrd_base
));
903 _FDT(fdt_setprop_cell(fdt
, chosen
, "linux,initrd-end",
904 spapr
->initrd_base
+ spapr
->initrd_size
));
906 if (spapr
->kernel_size
) {
907 uint64_t kprop
[2] = { cpu_to_be64(KERNEL_LOAD_ADDR
),
908 cpu_to_be64(spapr
->kernel_size
) };
910 _FDT(fdt_setprop(fdt
, chosen
, "qemu,boot-kernel",
911 &kprop
, sizeof(kprop
)));
912 if (spapr
->kernel_le
) {
913 _FDT(fdt_setprop(fdt
, chosen
, "qemu,boot-kernel-le", NULL
, 0));
917 _FDT((fdt_setprop_cell(fdt
, chosen
, "qemu,boot-menu", boot_menu
)));
919 _FDT(fdt_setprop_cell(fdt
, chosen
, "qemu,graphic-width", graphic_width
));
920 _FDT(fdt_setprop_cell(fdt
, chosen
, "qemu,graphic-height", graphic_height
));
921 _FDT(fdt_setprop_cell(fdt
, chosen
, "qemu,graphic-depth", graphic_depth
));
923 if (cb
&& bootlist
) {
926 for (i
= 0; i
< cb
; i
++) {
927 if (bootlist
[i
] == '\n') {
931 _FDT(fdt_setprop_string(fdt
, chosen
, "qemu,boot-list", bootlist
));
934 if (boot_device
&& strlen(boot_device
)) {
935 _FDT(fdt_setprop_string(fdt
, chosen
, "qemu,boot-device", boot_device
));
938 if (!spapr
->has_graphics
&& stdout_path
) {
939 _FDT(fdt_setprop_string(fdt
, chosen
, "linux,stdout-path", stdout_path
));
942 spapr_dt_ov5_platform_support(fdt
, chosen
);
948 static void spapr_dt_hypervisor(sPAPRMachineState
*spapr
, void *fdt
)
950 /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
951 * KVM to work under pHyp with some guest co-operation */
953 uint8_t hypercall
[16];
955 _FDT(hypervisor
= fdt_add_subnode(fdt
, 0, "hypervisor"));
956 /* indicate KVM hypercall interface */
957 _FDT(fdt_setprop_string(fdt
, hypervisor
, "compatible", "linux,kvm"));
958 if (kvmppc_has_cap_fixup_hcalls()) {
960 * Older KVM versions with older guest kernels were broken
961 * with the magic page, don't allow the guest to map it.
963 if (!kvmppc_get_hypercall(first_cpu
->env_ptr
, hypercall
,
964 sizeof(hypercall
))) {
965 _FDT(fdt_setprop(fdt
, hypervisor
, "hcall-instructions",
966 hypercall
, sizeof(hypercall
)));
971 static void *spapr_build_fdt(sPAPRMachineState
*spapr
,
975 MachineState
*machine
= MACHINE(qdev_get_machine());
976 MachineClass
*mc
= MACHINE_GET_CLASS(machine
);
977 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(machine
);
982 int smt
= kvmppc_smt_threads();
984 fdt
= g_malloc0(FDT_MAX_SIZE
);
985 _FDT((fdt_create_empty_tree(fdt
, FDT_MAX_SIZE
)));
988 _FDT(fdt_setprop_string(fdt
, 0, "device_type", "chrp"));
989 _FDT(fdt_setprop_string(fdt
, 0, "model", "IBM pSeries (emulated by qemu)"));
990 _FDT(fdt_setprop_string(fdt
, 0, "compatible", "qemu,pseries"));
993 * Add info to guest to indentify which host is it being run on
994 * and what is the uuid of the guest
996 if (kvmppc_get_host_model(&buf
)) {
997 _FDT(fdt_setprop_string(fdt
, 0, "host-model", buf
));
1000 if (kvmppc_get_host_serial(&buf
)) {
1001 _FDT(fdt_setprop_string(fdt
, 0, "host-serial", buf
));
1005 buf
= qemu_uuid_unparse_strdup(&qemu_uuid
);
1007 _FDT(fdt_setprop_string(fdt
, 0, "vm,uuid", buf
));
1008 if (qemu_uuid_set
) {
1009 _FDT(fdt_setprop_string(fdt
, 0, "system-id", buf
));
1013 if (qemu_get_vm_name()) {
1014 _FDT(fdt_setprop_string(fdt
, 0, "ibm,partition-name",
1015 qemu_get_vm_name()));
1018 _FDT(fdt_setprop_cell(fdt
, 0, "#address-cells", 2));
1019 _FDT(fdt_setprop_cell(fdt
, 0, "#size-cells", 2));
1021 /* /interrupt controller */
1022 spapr_dt_xics(DIV_ROUND_UP(max_cpus
* smt
, smp_threads
), fdt
, PHANDLE_XICP
);
1024 ret
= spapr_populate_memory(spapr
, fdt
);
1026 error_report("couldn't setup memory nodes in fdt");
1031 spapr_dt_vdevice(spapr
->vio_bus
, fdt
);
1033 if (object_resolve_path_type("", TYPE_SPAPR_RNG
, NULL
)) {
1034 ret
= spapr_rng_populate_dt(fdt
);
1036 error_report("could not set up rng device in the fdt");
1041 QLIST_FOREACH(phb
, &spapr
->phbs
, list
) {
1042 ret
= spapr_populate_pci_dt(phb
, PHANDLE_XICP
, fdt
);
1044 error_report("couldn't setup PCI devices in fdt");
1050 spapr_populate_cpus_dt_node(fdt
, spapr
);
1052 if (smc
->dr_lmb_enabled
) {
1053 _FDT(spapr_drc_populate_dt(fdt
, 0, NULL
, SPAPR_DR_CONNECTOR_TYPE_LMB
));
1056 if (mc
->has_hotpluggable_cpus
) {
1057 int offset
= fdt_path_offset(fdt
, "/cpus");
1058 ret
= spapr_drc_populate_dt(fdt
, offset
, NULL
,
1059 SPAPR_DR_CONNECTOR_TYPE_CPU
);
1061 error_report("Couldn't set up CPU DR device tree properties");
1066 /* /event-sources */
1067 spapr_dt_events(spapr
, fdt
);
1070 spapr_dt_rtas(spapr
, fdt
);
1073 spapr_dt_chosen(spapr
, fdt
);
1076 if (kvm_enabled()) {
1077 spapr_dt_hypervisor(spapr
, fdt
);
1080 /* Build memory reserve map */
1081 if (spapr
->kernel_size
) {
1082 _FDT((fdt_add_mem_rsv(fdt
, KERNEL_LOAD_ADDR
, spapr
->kernel_size
)));
1084 if (spapr
->initrd_size
) {
1085 _FDT((fdt_add_mem_rsv(fdt
, spapr
->initrd_base
, spapr
->initrd_size
)));
1088 /* ibm,client-architecture-support updates */
1089 ret
= spapr_dt_cas_updates(spapr
, fdt
, spapr
->ov5_cas
);
1091 error_report("couldn't setup CAS properties fdt");
1098 static uint64_t translate_kernel_address(void *opaque
, uint64_t addr
)
1100 return (addr
& 0x0fffffff) + KERNEL_LOAD_ADDR
;
1103 static void emulate_spapr_hypercall(PPCVirtualHypervisor
*vhyp
,
1106 CPUPPCState
*env
= &cpu
->env
;
1108 /* The TCG path should also be holding the BQL at this point */
1109 g_assert(qemu_mutex_iothread_locked());
1112 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1113 env
->gpr
[3] = H_PRIVILEGE
;
1115 env
->gpr
[3] = spapr_hypercall(cpu
, env
->gpr
[3], &env
->gpr
[4]);
1119 static uint64_t spapr_get_patbe(PPCVirtualHypervisor
*vhyp
)
1121 sPAPRMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1123 return spapr
->patb_entry
;
1126 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1127 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1128 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1129 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1130 #define DIRTY_HPTE(_hpte) ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1133 * Get the fd to access the kernel htab, re-opening it if necessary
1135 static int get_htab_fd(sPAPRMachineState
*spapr
)
1137 if (spapr
->htab_fd
>= 0) {
1138 return spapr
->htab_fd
;
1141 spapr
->htab_fd
= kvmppc_get_htab_fd(false);
1142 if (spapr
->htab_fd
< 0) {
1143 error_report("Unable to open fd for reading hash table from KVM: %s",
1147 return spapr
->htab_fd
;
1150 void close_htab_fd(sPAPRMachineState
*spapr
)
1152 if (spapr
->htab_fd
>= 0) {
1153 close(spapr
->htab_fd
);
1155 spapr
->htab_fd
= -1;
1158 static hwaddr
spapr_hpt_mask(PPCVirtualHypervisor
*vhyp
)
1160 sPAPRMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1162 return HTAB_SIZE(spapr
) / HASH_PTEG_SIZE_64
- 1;
1165 static const ppc_hash_pte64_t
*spapr_map_hptes(PPCVirtualHypervisor
*vhyp
,
1168 sPAPRMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1169 hwaddr pte_offset
= ptex
* HASH_PTE_SIZE_64
;
1173 * HTAB is controlled by KVM. Fetch into temporary buffer
1175 ppc_hash_pte64_t
*hptes
= g_malloc(n
* HASH_PTE_SIZE_64
);
1176 kvmppc_read_hptes(hptes
, ptex
, n
);
1181 * HTAB is controlled by QEMU. Just point to the internally
1184 return (const ppc_hash_pte64_t
*)(spapr
->htab
+ pte_offset
);
1187 static void spapr_unmap_hptes(PPCVirtualHypervisor
*vhyp
,
1188 const ppc_hash_pte64_t
*hptes
,
1191 sPAPRMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1194 g_free((void *)hptes
);
1197 /* Nothing to do for qemu managed HPT */
1200 static void spapr_store_hpte(PPCVirtualHypervisor
*vhyp
, hwaddr ptex
,
1201 uint64_t pte0
, uint64_t pte1
)
1203 sPAPRMachineState
*spapr
= SPAPR_MACHINE(vhyp
);
1204 hwaddr offset
= ptex
* HASH_PTE_SIZE_64
;
1207 kvmppc_write_hpte(ptex
, pte0
, pte1
);
1209 stq_p(spapr
->htab
+ offset
, pte0
);
1210 stq_p(spapr
->htab
+ offset
+ HASH_PTE_SIZE_64
/ 2, pte1
);
1214 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize
)
1218 /* We aim for a hash table of size 1/128 the size of RAM (rounded
1219 * up). The PAPR recommendation is actually 1/64 of RAM size, but
1220 * that's much more than is needed for Linux guests */
1221 shift
= ctz64(pow2ceil(ramsize
)) - 7;
1222 shift
= MAX(shift
, 18); /* Minimum architected size */
1223 shift
= MIN(shift
, 46); /* Maximum architected size */
1227 void spapr_free_hpt(sPAPRMachineState
*spapr
)
1229 g_free(spapr
->htab
);
1231 spapr
->htab_shift
= 0;
1232 close_htab_fd(spapr
);
1235 static void spapr_reallocate_hpt(sPAPRMachineState
*spapr
, int shift
,
1240 /* Clean up any HPT info from a previous boot */
1241 spapr_free_hpt(spapr
);
1243 rc
= kvmppc_reset_htab(shift
);
1245 /* kernel-side HPT needed, but couldn't allocate one */
1246 error_setg_errno(errp
, errno
,
1247 "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1249 /* This is almost certainly fatal, but if the caller really
1250 * wants to carry on with shift == 0, it's welcome to try */
1251 } else if (rc
> 0) {
1252 /* kernel-side HPT allocated */
1255 "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1259 spapr
->htab_shift
= shift
;
1262 /* kernel-side HPT not needed, allocate in userspace instead */
1263 size_t size
= 1ULL << shift
;
1266 spapr
->htab
= qemu_memalign(size
, size
);
1268 error_setg_errno(errp
, errno
,
1269 "Could not allocate HPT of order %d", shift
);
1273 memset(spapr
->htab
, 0, size
);
1274 spapr
->htab_shift
= shift
;
1276 for (i
= 0; i
< size
/ HASH_PTE_SIZE_64
; i
++) {
1277 DIRTY_HPTE(HPTE(spapr
->htab
, i
));
1282 void spapr_setup_hpt_and_vrma(sPAPRMachineState
*spapr
)
1284 spapr_reallocate_hpt(spapr
,
1285 spapr_hpt_shift_for_ramsize(MACHINE(spapr
)->maxram_size
),
1287 if (spapr
->vrma_adjust
) {
1288 spapr
->rma_size
= kvmppc_rma_size(spapr_node0_size(),
1291 /* We're setting up a hash table, so that means we're not radix */
1292 spapr
->patb_entry
= 0;
1295 static void find_unknown_sysbus_device(SysBusDevice
*sbdev
, void *opaque
)
1297 bool matched
= false;
1299 if (object_dynamic_cast(OBJECT(sbdev
), TYPE_SPAPR_PCI_HOST_BRIDGE
)) {
1304 error_report("Device %s is not supported by this machine yet.",
1305 qdev_fw_name(DEVICE(sbdev
)));
1310 static void ppc_spapr_reset(void)
1312 MachineState
*machine
= MACHINE(qdev_get_machine());
1313 sPAPRMachineState
*spapr
= SPAPR_MACHINE(machine
);
1314 PowerPCCPU
*first_ppc_cpu
;
1315 uint32_t rtas_limit
;
1316 hwaddr rtas_addr
, fdt_addr
;
1320 /* Check for unknown sysbus devices */
1321 foreach_dynamic_sysbus_device(find_unknown_sysbus_device
, NULL
);
1323 if (kvm_enabled() && kvmppc_has_cap_mmu_radix()) {
1324 /* If using KVM with radix mode available, VCPUs can be started
1325 * without a HPT because KVM will start them in radix mode.
1326 * Set the GR bit in PATB so that we know there is no HPT. */
1327 spapr
->patb_entry
= PATBE1_GR
;
1329 spapr
->patb_entry
= 0;
1330 spapr_setup_hpt_and_vrma(spapr
);
1333 qemu_devices_reset();
1336 * We place the device tree and RTAS just below either the top of the RMA,
1337 * or just below 2GB, whichever is lowere, so that it can be
1338 * processed with 32-bit real mode code if necessary
1340 rtas_limit
= MIN(spapr
->rma_size
, RTAS_MAX_ADDR
);
1341 rtas_addr
= rtas_limit
- RTAS_MAX_SIZE
;
1342 fdt_addr
= rtas_addr
- FDT_MAX_SIZE
;
1344 /* if this reset wasn't generated by CAS, we should reset our
1345 * negotiated options and start from scratch */
1346 if (!spapr
->cas_reboot
) {
1347 spapr_ovec_cleanup(spapr
->ov5_cas
);
1348 spapr
->ov5_cas
= spapr_ovec_new();
1351 fdt
= spapr_build_fdt(spapr
, rtas_addr
, spapr
->rtas_size
);
1353 spapr_load_rtas(spapr
, fdt
, rtas_addr
);
1357 /* Should only fail if we've built a corrupted tree */
1360 if (fdt_totalsize(fdt
) > FDT_MAX_SIZE
) {
1361 error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1362 fdt_totalsize(fdt
), FDT_MAX_SIZE
);
1367 qemu_fdt_dumpdtb(fdt
, fdt_totalsize(fdt
));
1368 cpu_physical_memory_write(fdt_addr
, fdt
, fdt_totalsize(fdt
));
1371 /* Set up the entry state */
1372 first_ppc_cpu
= POWERPC_CPU(first_cpu
);
1373 first_ppc_cpu
->env
.gpr
[3] = fdt_addr
;
1374 first_ppc_cpu
->env
.gpr
[5] = 0;
1375 first_cpu
->halted
= 0;
1376 first_ppc_cpu
->env
.nip
= SPAPR_ENTRY_POINT
;
1378 spapr
->cas_reboot
= false;
1381 static void spapr_create_nvram(sPAPRMachineState
*spapr
)
1383 DeviceState
*dev
= qdev_create(&spapr
->vio_bus
->bus
, "spapr-nvram");
1384 DriveInfo
*dinfo
= drive_get(IF_PFLASH
, 0, 0);
1387 qdev_prop_set_drive(dev
, "drive", blk_by_legacy_dinfo(dinfo
),
1391 qdev_init_nofail(dev
);
1393 spapr
->nvram
= (struct sPAPRNVRAM
*)dev
;
1396 static void spapr_rtc_create(sPAPRMachineState
*spapr
)
1398 object_initialize(&spapr
->rtc
, sizeof(spapr
->rtc
), TYPE_SPAPR_RTC
);
1399 object_property_add_child(OBJECT(spapr
), "rtc", OBJECT(&spapr
->rtc
),
1401 object_property_set_bool(OBJECT(&spapr
->rtc
), true, "realized",
1403 object_property_add_alias(OBJECT(spapr
), "rtc-time", OBJECT(&spapr
->rtc
),
1404 "date", &error_fatal
);
1407 /* Returns whether we want to use VGA or not */
1408 static bool spapr_vga_init(PCIBus
*pci_bus
, Error
**errp
)
1410 switch (vga_interface_type
) {
1417 return pci_vga_init(pci_bus
) != NULL
;
1420 "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1425 static int spapr_post_load(void *opaque
, int version_id
)
1427 sPAPRMachineState
*spapr
= (sPAPRMachineState
*)opaque
;
1430 if (!object_dynamic_cast(OBJECT(spapr
->ics
), TYPE_ICS_KVM
)) {
1433 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
1434 icp_resend(ICP(cpu
->intc
));
1438 /* In earlier versions, there was no separate qdev for the PAPR
1439 * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1440 * So when migrating from those versions, poke the incoming offset
1441 * value into the RTC device */
1442 if (version_id
< 3) {
1443 err
= spapr_rtc_import_offset(&spapr
->rtc
, spapr
->rtc_offset
);
1449 static bool version_before_3(void *opaque
, int version_id
)
1451 return version_id
< 3;
1454 static bool spapr_ov5_cas_needed(void *opaque
)
1456 sPAPRMachineState
*spapr
= opaque
;
1457 sPAPROptionVector
*ov5_mask
= spapr_ovec_new();
1458 sPAPROptionVector
*ov5_legacy
= spapr_ovec_new();
1459 sPAPROptionVector
*ov5_removed
= spapr_ovec_new();
1462 /* Prior to the introduction of sPAPROptionVector, we had two option
1463 * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1464 * Both of these options encode machine topology into the device-tree
1465 * in such a way that the now-booted OS should still be able to interact
1466 * appropriately with QEMU regardless of what options were actually
1467 * negotiatied on the source side.
1469 * As such, we can avoid migrating the CAS-negotiated options if these
1470 * are the only options available on the current machine/platform.
1471 * Since these are the only options available for pseries-2.7 and
1472 * earlier, this allows us to maintain old->new/new->old migration
1475 * For QEMU 2.8+, there are additional CAS-negotiatable options available
1476 * via default pseries-2.8 machines and explicit command-line parameters.
1477 * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1478 * of the actual CAS-negotiated values to continue working properly. For
1479 * example, availability of memory unplug depends on knowing whether
1480 * OV5_HP_EVT was negotiated via CAS.
1482 * Thus, for any cases where the set of available CAS-negotiatable
1483 * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1484 * include the CAS-negotiated options in the migration stream.
1486 spapr_ovec_set(ov5_mask
, OV5_FORM1_AFFINITY
);
1487 spapr_ovec_set(ov5_mask
, OV5_DRCONF_MEMORY
);
1489 /* spapr_ovec_diff returns true if bits were removed. we avoid using
1490 * the mask itself since in the future it's possible "legacy" bits may be
1491 * removed via machine options, which could generate a false positive
1492 * that breaks migration.
1494 spapr_ovec_intersect(ov5_legacy
, spapr
->ov5
, ov5_mask
);
1495 cas_needed
= spapr_ovec_diff(ov5_removed
, spapr
->ov5
, ov5_legacy
);
1497 spapr_ovec_cleanup(ov5_mask
);
1498 spapr_ovec_cleanup(ov5_legacy
);
1499 spapr_ovec_cleanup(ov5_removed
);
1504 static const VMStateDescription vmstate_spapr_ov5_cas
= {
1505 .name
= "spapr_option_vector_ov5_cas",
1507 .minimum_version_id
= 1,
1508 .needed
= spapr_ov5_cas_needed
,
1509 .fields
= (VMStateField
[]) {
1510 VMSTATE_STRUCT_POINTER_V(ov5_cas
, sPAPRMachineState
, 1,
1511 vmstate_spapr_ovec
, sPAPROptionVector
),
1512 VMSTATE_END_OF_LIST()
1516 static bool spapr_patb_entry_needed(void *opaque
)
1518 sPAPRMachineState
*spapr
= opaque
;
1520 return !!spapr
->patb_entry
;
1523 static const VMStateDescription vmstate_spapr_patb_entry
= {
1524 .name
= "spapr_patb_entry",
1526 .minimum_version_id
= 1,
1527 .needed
= spapr_patb_entry_needed
,
1528 .fields
= (VMStateField
[]) {
1529 VMSTATE_UINT64(patb_entry
, sPAPRMachineState
),
1530 VMSTATE_END_OF_LIST()
1534 static const VMStateDescription vmstate_spapr
= {
1537 .minimum_version_id
= 1,
1538 .post_load
= spapr_post_load
,
1539 .fields
= (VMStateField
[]) {
1540 /* used to be @next_irq */
1541 VMSTATE_UNUSED_BUFFER(version_before_3
, 0, 4),
1544 VMSTATE_UINT64_TEST(rtc_offset
, sPAPRMachineState
, version_before_3
),
1546 VMSTATE_PPC_TIMEBASE_V(tb
, sPAPRMachineState
, 2),
1547 VMSTATE_END_OF_LIST()
1549 .subsections
= (const VMStateDescription
*[]) {
1550 &vmstate_spapr_ov5_cas
,
1551 &vmstate_spapr_patb_entry
,
1556 static int htab_save_setup(QEMUFile
*f
, void *opaque
)
1558 sPAPRMachineState
*spapr
= opaque
;
1560 /* "Iteration" header */
1561 qemu_put_be32(f
, spapr
->htab_shift
);
1564 spapr
->htab_save_index
= 0;
1565 spapr
->htab_first_pass
= true;
1567 assert(kvm_enabled());
1574 static void htab_save_first_pass(QEMUFile
*f
, sPAPRMachineState
*spapr
,
1577 bool has_timeout
= max_ns
!= -1;
1578 int htabslots
= HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
;
1579 int index
= spapr
->htab_save_index
;
1580 int64_t starttime
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
1582 assert(spapr
->htab_first_pass
);
1587 /* Consume invalid HPTEs */
1588 while ((index
< htabslots
)
1589 && !HPTE_VALID(HPTE(spapr
->htab
, index
))) {
1590 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
1594 /* Consume valid HPTEs */
1596 while ((index
< htabslots
) && (index
- chunkstart
< USHRT_MAX
)
1597 && HPTE_VALID(HPTE(spapr
->htab
, index
))) {
1598 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
1602 if (index
> chunkstart
) {
1603 int n_valid
= index
- chunkstart
;
1605 qemu_put_be32(f
, chunkstart
);
1606 qemu_put_be16(f
, n_valid
);
1607 qemu_put_be16(f
, 0);
1608 qemu_put_buffer(f
, HPTE(spapr
->htab
, chunkstart
),
1609 HASH_PTE_SIZE_64
* n_valid
);
1612 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - starttime
) > max_ns
) {
1616 } while ((index
< htabslots
) && !qemu_file_rate_limit(f
));
1618 if (index
>= htabslots
) {
1619 assert(index
== htabslots
);
1621 spapr
->htab_first_pass
= false;
1623 spapr
->htab_save_index
= index
;
1626 static int htab_save_later_pass(QEMUFile
*f
, sPAPRMachineState
*spapr
,
1629 bool final
= max_ns
< 0;
1630 int htabslots
= HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
;
1631 int examined
= 0, sent
= 0;
1632 int index
= spapr
->htab_save_index
;
1633 int64_t starttime
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
1635 assert(!spapr
->htab_first_pass
);
1638 int chunkstart
, invalidstart
;
1640 /* Consume non-dirty HPTEs */
1641 while ((index
< htabslots
)
1642 && !HPTE_DIRTY(HPTE(spapr
->htab
, index
))) {
1648 /* Consume valid dirty HPTEs */
1649 while ((index
< htabslots
) && (index
- chunkstart
< USHRT_MAX
)
1650 && HPTE_DIRTY(HPTE(spapr
->htab
, index
))
1651 && HPTE_VALID(HPTE(spapr
->htab
, index
))) {
1652 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
1657 invalidstart
= index
;
1658 /* Consume invalid dirty HPTEs */
1659 while ((index
< htabslots
) && (index
- invalidstart
< USHRT_MAX
)
1660 && HPTE_DIRTY(HPTE(spapr
->htab
, index
))
1661 && !HPTE_VALID(HPTE(spapr
->htab
, index
))) {
1662 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
1667 if (index
> chunkstart
) {
1668 int n_valid
= invalidstart
- chunkstart
;
1669 int n_invalid
= index
- invalidstart
;
1671 qemu_put_be32(f
, chunkstart
);
1672 qemu_put_be16(f
, n_valid
);
1673 qemu_put_be16(f
, n_invalid
);
1674 qemu_put_buffer(f
, HPTE(spapr
->htab
, chunkstart
),
1675 HASH_PTE_SIZE_64
* n_valid
);
1676 sent
+= index
- chunkstart
;
1678 if (!final
&& (qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - starttime
) > max_ns
) {
1683 if (examined
>= htabslots
) {
1687 if (index
>= htabslots
) {
1688 assert(index
== htabslots
);
1691 } while ((examined
< htabslots
) && (!qemu_file_rate_limit(f
) || final
));
1693 if (index
>= htabslots
) {
1694 assert(index
== htabslots
);
1698 spapr
->htab_save_index
= index
;
1700 return (examined
>= htabslots
) && (sent
== 0) ? 1 : 0;
1703 #define MAX_ITERATION_NS 5000000 /* 5 ms */
1704 #define MAX_KVM_BUF_SIZE 2048
1706 static int htab_save_iterate(QEMUFile
*f
, void *opaque
)
1708 sPAPRMachineState
*spapr
= opaque
;
1712 /* Iteration header */
1713 qemu_put_be32(f
, 0);
1716 assert(kvm_enabled());
1718 fd
= get_htab_fd(spapr
);
1723 rc
= kvmppc_save_htab(f
, fd
, MAX_KVM_BUF_SIZE
, MAX_ITERATION_NS
);
1727 } else if (spapr
->htab_first_pass
) {
1728 htab_save_first_pass(f
, spapr
, MAX_ITERATION_NS
);
1730 rc
= htab_save_later_pass(f
, spapr
, MAX_ITERATION_NS
);
1734 qemu_put_be32(f
, 0);
1735 qemu_put_be16(f
, 0);
1736 qemu_put_be16(f
, 0);
1741 static int htab_save_complete(QEMUFile
*f
, void *opaque
)
1743 sPAPRMachineState
*spapr
= opaque
;
1746 /* Iteration header */
1747 qemu_put_be32(f
, 0);
1752 assert(kvm_enabled());
1754 fd
= get_htab_fd(spapr
);
1759 rc
= kvmppc_save_htab(f
, fd
, MAX_KVM_BUF_SIZE
, -1);
1764 if (spapr
->htab_first_pass
) {
1765 htab_save_first_pass(f
, spapr
, -1);
1767 htab_save_later_pass(f
, spapr
, -1);
1771 qemu_put_be32(f
, 0);
1772 qemu_put_be16(f
, 0);
1773 qemu_put_be16(f
, 0);
1778 static int htab_load(QEMUFile
*f
, void *opaque
, int version_id
)
1780 sPAPRMachineState
*spapr
= opaque
;
1781 uint32_t section_hdr
;
1784 if (version_id
< 1 || version_id
> 1) {
1785 error_report("htab_load() bad version");
1789 section_hdr
= qemu_get_be32(f
);
1792 Error
*local_err
= NULL
;
1794 /* First section gives the htab size */
1795 spapr_reallocate_hpt(spapr
, section_hdr
, &local_err
);
1797 error_report_err(local_err
);
1804 assert(kvm_enabled());
1806 fd
= kvmppc_get_htab_fd(true);
1808 error_report("Unable to open fd to restore KVM hash table: %s",
1815 uint16_t n_valid
, n_invalid
;
1817 index
= qemu_get_be32(f
);
1818 n_valid
= qemu_get_be16(f
);
1819 n_invalid
= qemu_get_be16(f
);
1821 if ((index
== 0) && (n_valid
== 0) && (n_invalid
== 0)) {
1826 if ((index
+ n_valid
+ n_invalid
) >
1827 (HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
)) {
1828 /* Bad index in stream */
1830 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1831 index
, n_valid
, n_invalid
, spapr
->htab_shift
);
1837 qemu_get_buffer(f
, HPTE(spapr
->htab
, index
),
1838 HASH_PTE_SIZE_64
* n_valid
);
1841 memset(HPTE(spapr
->htab
, index
+ n_valid
), 0,
1842 HASH_PTE_SIZE_64
* n_invalid
);
1849 rc
= kvmppc_load_htab_chunk(f
, fd
, index
, n_valid
, n_invalid
);
1864 static void htab_cleanup(void *opaque
)
1866 sPAPRMachineState
*spapr
= opaque
;
1868 close_htab_fd(spapr
);
1871 static SaveVMHandlers savevm_htab_handlers
= {
1872 .save_live_setup
= htab_save_setup
,
1873 .save_live_iterate
= htab_save_iterate
,
1874 .save_live_complete_precopy
= htab_save_complete
,
1875 .cleanup
= htab_cleanup
,
1876 .load_state
= htab_load
,
1879 static void spapr_boot_set(void *opaque
, const char *boot_device
,
1882 MachineState
*machine
= MACHINE(qdev_get_machine());
1883 machine
->boot_order
= g_strdup(boot_device
);
1887 * Reset routine for LMB DR devices.
1889 * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1890 * routine. Reset for PCI DR devices will be handled by PHB reset routine
1891 * when it walks all its children devices. LMB devices reset occurs
1892 * as part of spapr_ppc_reset().
1894 static void spapr_drc_reset(void *opaque
)
1896 sPAPRDRConnector
*drc
= opaque
;
1897 DeviceState
*d
= DEVICE(drc
);
1904 static void spapr_create_lmb_dr_connectors(sPAPRMachineState
*spapr
)
1906 MachineState
*machine
= MACHINE(spapr
);
1907 uint64_t lmb_size
= SPAPR_MEMORY_BLOCK_SIZE
;
1908 uint32_t nr_lmbs
= (machine
->maxram_size
- machine
->ram_size
)/lmb_size
;
1911 for (i
= 0; i
< nr_lmbs
; i
++) {
1912 sPAPRDRConnector
*drc
;
1915 addr
= i
* lmb_size
+ spapr
->hotplug_memory
.base
;
1916 drc
= spapr_dr_connector_new(OBJECT(spapr
), TYPE_SPAPR_DRC_LMB
,
1918 qemu_register_reset(spapr_drc_reset
, drc
);
1923 * If RAM size, maxmem size and individual node mem sizes aren't aligned
1924 * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1925 * since we can't support such unaligned sizes with DRCONF_MEMORY.
1927 static void spapr_validate_node_memory(MachineState
*machine
, Error
**errp
)
1931 if (machine
->ram_size
% SPAPR_MEMORY_BLOCK_SIZE
) {
1932 error_setg(errp
, "Memory size 0x" RAM_ADDR_FMT
1933 " is not aligned to %llu MiB",
1935 SPAPR_MEMORY_BLOCK_SIZE
/ M_BYTE
);
1939 if (machine
->maxram_size
% SPAPR_MEMORY_BLOCK_SIZE
) {
1940 error_setg(errp
, "Maximum memory size 0x" RAM_ADDR_FMT
1941 " is not aligned to %llu MiB",
1943 SPAPR_MEMORY_BLOCK_SIZE
/ M_BYTE
);
1947 for (i
= 0; i
< nb_numa_nodes
; i
++) {
1948 if (numa_info
[i
].node_mem
% SPAPR_MEMORY_BLOCK_SIZE
) {
1950 "Node %d memory size 0x%" PRIx64
1951 " is not aligned to %llu MiB",
1952 i
, numa_info
[i
].node_mem
,
1953 SPAPR_MEMORY_BLOCK_SIZE
/ M_BYTE
);
1959 /* find cpu slot in machine->possible_cpus by core_id */
1960 static CPUArchId
*spapr_find_cpu_slot(MachineState
*ms
, uint32_t id
, int *idx
)
1962 int index
= id
/ smp_threads
;
1964 if (index
>= ms
->possible_cpus
->len
) {
1970 return &ms
->possible_cpus
->cpus
[index
];
1973 static void spapr_init_cpus(sPAPRMachineState
*spapr
)
1975 MachineState
*machine
= MACHINE(spapr
);
1976 MachineClass
*mc
= MACHINE_GET_CLASS(machine
);
1977 char *type
= spapr_get_cpu_core_type(machine
->cpu_model
);
1978 int smt
= kvmppc_smt_threads();
1979 const CPUArchIdList
*possible_cpus
;
1980 int boot_cores_nr
= smp_cpus
/ smp_threads
;
1984 error_report("Unable to find sPAPR CPU Core definition");
1988 possible_cpus
= mc
->possible_cpu_arch_ids(machine
);
1989 if (mc
->has_hotpluggable_cpus
) {
1990 if (smp_cpus
% smp_threads
) {
1991 error_report("smp_cpus (%u) must be multiple of threads (%u)",
1992 smp_cpus
, smp_threads
);
1995 if (max_cpus
% smp_threads
) {
1996 error_report("max_cpus (%u) must be multiple of threads (%u)",
1997 max_cpus
, smp_threads
);
2001 if (max_cpus
!= smp_cpus
) {
2002 error_report("This machine version does not support CPU hotplug");
2005 boot_cores_nr
= possible_cpus
->len
;
2008 for (i
= 0; i
< possible_cpus
->len
; i
++) {
2009 int core_id
= i
* smp_threads
;
2011 if (mc
->has_hotpluggable_cpus
) {
2012 sPAPRDRConnector
*drc
=
2013 spapr_dr_connector_new(OBJECT(spapr
), TYPE_SPAPR_DRC_CPU
,
2014 (core_id
/ smp_threads
) * smt
);
2016 qemu_register_reset(spapr_drc_reset
, drc
);
2019 if (i
< boot_cores_nr
) {
2020 Object
*core
= object_new(type
);
2021 int nr_threads
= smp_threads
;
2023 /* Handle the partially filled core for older machine types */
2024 if ((i
+ 1) * smp_threads
>= smp_cpus
) {
2025 nr_threads
= smp_cpus
- i
* smp_threads
;
2028 object_property_set_int(core
, nr_threads
, "nr-threads",
2030 object_property_set_int(core
, core_id
, CPU_CORE_PROP_CORE_ID
,
2032 object_property_set_bool(core
, true, "realized", &error_fatal
);
2038 /* pSeries LPAR / sPAPR hardware init */
2039 static void ppc_spapr_init(MachineState
*machine
)
2041 sPAPRMachineState
*spapr
= SPAPR_MACHINE(machine
);
2042 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(machine
);
2043 const char *kernel_filename
= machine
->kernel_filename
;
2044 const char *initrd_filename
= machine
->initrd_filename
;
2047 MemoryRegion
*sysmem
= get_system_memory();
2048 MemoryRegion
*ram
= g_new(MemoryRegion
, 1);
2049 MemoryRegion
*rma_region
;
2051 hwaddr rma_alloc_size
;
2052 hwaddr node0_size
= spapr_node0_size();
2053 long load_limit
, fw_size
;
2056 msi_nonbroken
= true;
2058 QLIST_INIT(&spapr
->phbs
);
2059 QTAILQ_INIT(&spapr
->pending_dimm_unplugs
);
2061 /* Allocate RMA if necessary */
2062 rma_alloc_size
= kvmppc_alloc_rma(&rma
);
2064 if (rma_alloc_size
== -1) {
2065 error_report("Unable to create RMA");
2069 if (rma_alloc_size
&& (rma_alloc_size
< node0_size
)) {
2070 spapr
->rma_size
= rma_alloc_size
;
2072 spapr
->rma_size
= node0_size
;
2074 /* With KVM, we don't actually know whether KVM supports an
2075 * unbounded RMA (PR KVM) or is limited by the hash table size
2076 * (HV KVM using VRMA), so we always assume the latter
2078 * In that case, we also limit the initial allocations for RTAS
2079 * etc... to 256M since we have no way to know what the VRMA size
2080 * is going to be as it depends on the size of the hash table
2081 * isn't determined yet.
2083 if (kvm_enabled()) {
2084 spapr
->vrma_adjust
= 1;
2085 spapr
->rma_size
= MIN(spapr
->rma_size
, 0x10000000);
2088 /* Actually we don't support unbounded RMA anymore since we
2089 * added proper emulation of HV mode. The max we can get is
2090 * 16G which also happens to be what we configure for PAPR
2091 * mode so make sure we don't do anything bigger than that
2093 spapr
->rma_size
= MIN(spapr
->rma_size
, 0x400000000ull
);
2096 if (spapr
->rma_size
> node0_size
) {
2097 error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx
")",
2102 /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2103 load_limit
= MIN(spapr
->rma_size
, RTAS_MAX_ADDR
) - FW_OVERHEAD
;
2105 /* Set up Interrupt Controller before we create the VCPUs */
2106 xics_system_init(machine
, XICS_IRQS_SPAPR
, &error_fatal
);
2108 /* Set up containers for ibm,client-set-architecture negotiated options */
2109 spapr
->ov5
= spapr_ovec_new();
2110 spapr
->ov5_cas
= spapr_ovec_new();
2112 if (smc
->dr_lmb_enabled
) {
2113 spapr_ovec_set(spapr
->ov5
, OV5_DRCONF_MEMORY
);
2114 spapr_validate_node_memory(machine
, &error_fatal
);
2117 spapr_ovec_set(spapr
->ov5
, OV5_FORM1_AFFINITY
);
2118 if (!kvm_enabled() || kvmppc_has_cap_mmu_radix()) {
2119 /* KVM and TCG always allow GTSE with radix... */
2120 spapr_ovec_set(spapr
->ov5
, OV5_MMU_RADIX_GTSE
);
2122 /* ... but not with hash (currently). */
2124 /* advertise support for dedicated HP event source to guests */
2125 if (spapr
->use_hotplug_event_source
) {
2126 spapr_ovec_set(spapr
->ov5
, OV5_HP_EVT
);
2130 if (machine
->cpu_model
== NULL
) {
2131 machine
->cpu_model
= kvm_enabled() ? "host" : smc
->tcg_default_cpu
;
2134 ppc_cpu_parse_features(machine
->cpu_model
);
2136 spapr_init_cpus(spapr
);
2138 if (kvm_enabled()) {
2139 /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2140 kvmppc_enable_logical_ci_hcalls();
2141 kvmppc_enable_set_mode_hcall();
2143 /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2144 kvmppc_enable_clear_ref_mod_hcalls();
2148 memory_region_allocate_system_memory(ram
, NULL
, "ppc_spapr.ram",
2150 memory_region_add_subregion(sysmem
, 0, ram
);
2152 if (rma_alloc_size
&& rma
) {
2153 rma_region
= g_new(MemoryRegion
, 1);
2154 memory_region_init_ram_ptr(rma_region
, NULL
, "ppc_spapr.rma",
2155 rma_alloc_size
, rma
);
2156 vmstate_register_ram_global(rma_region
);
2157 memory_region_add_subregion(sysmem
, 0, rma_region
);
2160 /* initialize hotplug memory address space */
2161 if (machine
->ram_size
< machine
->maxram_size
) {
2162 ram_addr_t hotplug_mem_size
= machine
->maxram_size
- machine
->ram_size
;
2164 * Limit the number of hotpluggable memory slots to half the number
2165 * slots that KVM supports, leaving the other half for PCI and other
2166 * devices. However ensure that number of slots doesn't drop below 32.
2168 int max_memslots
= kvm_enabled() ? kvm_get_max_memslots() / 2 :
2169 SPAPR_MAX_RAM_SLOTS
;
2171 if (max_memslots
< SPAPR_MAX_RAM_SLOTS
) {
2172 max_memslots
= SPAPR_MAX_RAM_SLOTS
;
2174 if (machine
->ram_slots
> max_memslots
) {
2175 error_report("Specified number of memory slots %"
2176 PRIu64
" exceeds max supported %d",
2177 machine
->ram_slots
, max_memslots
);
2181 spapr
->hotplug_memory
.base
= ROUND_UP(machine
->ram_size
,
2182 SPAPR_HOTPLUG_MEM_ALIGN
);
2183 memory_region_init(&spapr
->hotplug_memory
.mr
, OBJECT(spapr
),
2184 "hotplug-memory", hotplug_mem_size
);
2185 memory_region_add_subregion(sysmem
, spapr
->hotplug_memory
.base
,
2186 &spapr
->hotplug_memory
.mr
);
2189 if (smc
->dr_lmb_enabled
) {
2190 spapr_create_lmb_dr_connectors(spapr
);
2193 filename
= qemu_find_file(QEMU_FILE_TYPE_BIOS
, "spapr-rtas.bin");
2195 error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2198 spapr
->rtas_size
= get_image_size(filename
);
2199 if (spapr
->rtas_size
< 0) {
2200 error_report("Could not get size of LPAR rtas '%s'", filename
);
2203 spapr
->rtas_blob
= g_malloc(spapr
->rtas_size
);
2204 if (load_image_size(filename
, spapr
->rtas_blob
, spapr
->rtas_size
) < 0) {
2205 error_report("Could not load LPAR rtas '%s'", filename
);
2208 if (spapr
->rtas_size
> RTAS_MAX_SIZE
) {
2209 error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2210 (size_t)spapr
->rtas_size
, RTAS_MAX_SIZE
);
2215 /* Set up RTAS event infrastructure */
2216 spapr_events_init(spapr
);
2218 /* Set up the RTC RTAS interfaces */
2219 spapr_rtc_create(spapr
);
2221 /* Set up VIO bus */
2222 spapr
->vio_bus
= spapr_vio_bus_init();
2224 for (i
= 0; i
< MAX_SERIAL_PORTS
; i
++) {
2225 if (serial_hds
[i
]) {
2226 spapr_vty_create(spapr
->vio_bus
, serial_hds
[i
]);
2230 /* We always have at least the nvram device on VIO */
2231 spapr_create_nvram(spapr
);
2234 spapr_pci_rtas_init();
2236 phb
= spapr_create_phb(spapr
, 0);
2238 for (i
= 0; i
< nb_nics
; i
++) {
2239 NICInfo
*nd
= &nd_table
[i
];
2242 nd
->model
= g_strdup("ibmveth");
2245 if (strcmp(nd
->model
, "ibmveth") == 0) {
2246 spapr_vlan_create(spapr
->vio_bus
, nd
);
2248 pci_nic_init_nofail(&nd_table
[i
], phb
->bus
, nd
->model
, NULL
);
2252 for (i
= 0; i
<= drive_get_max_bus(IF_SCSI
); i
++) {
2253 spapr_vscsi_create(spapr
->vio_bus
);
2257 if (spapr_vga_init(phb
->bus
, &error_fatal
)) {
2258 spapr
->has_graphics
= true;
2259 machine
->usb
|= defaults_enabled() && !machine
->usb_disabled
;
2263 if (smc
->use_ohci_by_default
) {
2264 pci_create_simple(phb
->bus
, -1, "pci-ohci");
2266 pci_create_simple(phb
->bus
, -1, "nec-usb-xhci");
2269 if (spapr
->has_graphics
) {
2270 USBBus
*usb_bus
= usb_bus_find(-1);
2272 usb_create_simple(usb_bus
, "usb-kbd");
2273 usb_create_simple(usb_bus
, "usb-mouse");
2277 if (spapr
->rma_size
< (MIN_RMA_SLOF
<< 20)) {
2279 "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2284 if (kernel_filename
) {
2285 uint64_t lowaddr
= 0;
2287 spapr
->kernel_size
= load_elf(kernel_filename
, translate_kernel_address
,
2288 NULL
, NULL
, &lowaddr
, NULL
, 1,
2289 PPC_ELF_MACHINE
, 0, 0);
2290 if (spapr
->kernel_size
== ELF_LOAD_WRONG_ENDIAN
) {
2291 spapr
->kernel_size
= load_elf(kernel_filename
,
2292 translate_kernel_address
, NULL
, NULL
,
2293 &lowaddr
, NULL
, 0, PPC_ELF_MACHINE
,
2295 spapr
->kernel_le
= spapr
->kernel_size
> 0;
2297 if (spapr
->kernel_size
< 0) {
2298 error_report("error loading %s: %s", kernel_filename
,
2299 load_elf_strerror(spapr
->kernel_size
));
2304 if (initrd_filename
) {
2305 /* Try to locate the initrd in the gap between the kernel
2306 * and the firmware. Add a bit of space just in case
2308 spapr
->initrd_base
= (KERNEL_LOAD_ADDR
+ spapr
->kernel_size
2309 + 0x1ffff) & ~0xffff;
2310 spapr
->initrd_size
= load_image_targphys(initrd_filename
,
2313 - spapr
->initrd_base
);
2314 if (spapr
->initrd_size
< 0) {
2315 error_report("could not load initial ram disk '%s'",
2322 if (bios_name
== NULL
) {
2323 bios_name
= FW_FILE_NAME
;
2325 filename
= qemu_find_file(QEMU_FILE_TYPE_BIOS
, bios_name
);
2327 error_report("Could not find LPAR firmware '%s'", bios_name
);
2330 fw_size
= load_image_targphys(filename
, 0, FW_MAX_SIZE
);
2332 error_report("Could not load LPAR firmware '%s'", filename
);
2337 /* FIXME: Should register things through the MachineState's qdev
2338 * interface, this is a legacy from the sPAPREnvironment structure
2339 * which predated MachineState but had a similar function */
2340 vmstate_register(NULL
, 0, &vmstate_spapr
, spapr
);
2341 register_savevm_live(NULL
, "spapr/htab", -1, 1,
2342 &savevm_htab_handlers
, spapr
);
2344 qemu_register_boot_set(spapr_boot_set
, spapr
);
2346 if (kvm_enabled()) {
2347 /* to stop and start vmclock */
2348 qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change
,
2351 kvmppc_spapr_enable_inkernel_multitce();
2355 static int spapr_kvm_type(const char *vm_type
)
2361 if (!strcmp(vm_type
, "HV")) {
2365 if (!strcmp(vm_type
, "PR")) {
2369 error_report("Unknown kvm-type specified '%s'", vm_type
);
2374 * Implementation of an interface to adjust firmware path
2375 * for the bootindex property handling.
2377 static char *spapr_get_fw_dev_path(FWPathProvider
*p
, BusState
*bus
,
2380 #define CAST(type, obj, name) \
2381 ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2382 SCSIDevice
*d
= CAST(SCSIDevice
, dev
, TYPE_SCSI_DEVICE
);
2383 sPAPRPHBState
*phb
= CAST(sPAPRPHBState
, dev
, TYPE_SPAPR_PCI_HOST_BRIDGE
);
2384 VHostSCSICommon
*vsc
= CAST(VHostSCSICommon
, dev
, TYPE_VHOST_SCSI_COMMON
);
2387 void *spapr
= CAST(void, bus
->parent
, "spapr-vscsi");
2388 VirtIOSCSI
*virtio
= CAST(VirtIOSCSI
, bus
->parent
, TYPE_VIRTIO_SCSI
);
2389 USBDevice
*usb
= CAST(USBDevice
, bus
->parent
, TYPE_USB_DEVICE
);
2393 * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2394 * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2395 * in the top 16 bits of the 64-bit LUN
2397 unsigned id
= 0x8000 | (d
->id
<< 8) | d
->lun
;
2398 return g_strdup_printf("%s@%"PRIX64
, qdev_fw_name(dev
),
2399 (uint64_t)id
<< 48);
2400 } else if (virtio
) {
2402 * We use SRP luns of the form 01000000 | (target << 8) | lun
2403 * in the top 32 bits of the 64-bit LUN
2404 * Note: the quote above is from SLOF and it is wrong,
2405 * the actual binding is:
2406 * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2408 unsigned id
= 0x1000000 | (d
->id
<< 16) | d
->lun
;
2409 return g_strdup_printf("%s@%"PRIX64
, qdev_fw_name(dev
),
2410 (uint64_t)id
<< 32);
2413 * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2414 * in the top 32 bits of the 64-bit LUN
2416 unsigned usb_port
= atoi(usb
->port
->path
);
2417 unsigned id
= 0x1000000 | (usb_port
<< 16) | d
->lun
;
2418 return g_strdup_printf("%s@%"PRIX64
, qdev_fw_name(dev
),
2419 (uint64_t)id
<< 32);
2424 * SLOF probes the USB devices, and if it recognizes that the device is a
2425 * storage device, it changes its name to "storage" instead of "usb-host",
2426 * and additionally adds a child node for the SCSI LUN, so the correct
2427 * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2429 if (strcmp("usb-host", qdev_fw_name(dev
)) == 0) {
2430 USBDevice
*usbdev
= CAST(USBDevice
, dev
, TYPE_USB_DEVICE
);
2431 if (usb_host_dev_is_scsi_storage(usbdev
)) {
2432 return g_strdup_printf("storage@%s/disk", usbdev
->port
->path
);
2437 /* Replace "pci" with "pci@800000020000000" */
2438 return g_strdup_printf("pci@%"PRIX64
, phb
->buid
);
2442 /* Same logic as virtio above */
2443 unsigned id
= 0x1000000 | (vsc
->target
<< 16) | vsc
->lun
;
2444 return g_strdup_printf("disk@%"PRIX64
, (uint64_t)id
<< 32);
2447 if (g_str_equal("pci-bridge", qdev_fw_name(dev
))) {
2448 /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2449 PCIDevice
*pcidev
= CAST(PCIDevice
, dev
, TYPE_PCI_DEVICE
);
2450 return g_strdup_printf("pci@%x", PCI_SLOT(pcidev
->devfn
));
2456 static char *spapr_get_kvm_type(Object
*obj
, Error
**errp
)
2458 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
2460 return g_strdup(spapr
->kvm_type
);
2463 static void spapr_set_kvm_type(Object
*obj
, const char *value
, Error
**errp
)
2465 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
2467 g_free(spapr
->kvm_type
);
2468 spapr
->kvm_type
= g_strdup(value
);
2471 static bool spapr_get_modern_hotplug_events(Object
*obj
, Error
**errp
)
2473 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
2475 return spapr
->use_hotplug_event_source
;
2478 static void spapr_set_modern_hotplug_events(Object
*obj
, bool value
,
2481 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
2483 spapr
->use_hotplug_event_source
= value
;
2486 static void spapr_machine_initfn(Object
*obj
)
2488 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
2490 spapr
->htab_fd
= -1;
2491 spapr
->use_hotplug_event_source
= true;
2492 object_property_add_str(obj
, "kvm-type",
2493 spapr_get_kvm_type
, spapr_set_kvm_type
, NULL
);
2494 object_property_set_description(obj
, "kvm-type",
2495 "Specifies the KVM virtualization mode (HV, PR)",
2497 object_property_add_bool(obj
, "modern-hotplug-events",
2498 spapr_get_modern_hotplug_events
,
2499 spapr_set_modern_hotplug_events
,
2501 object_property_set_description(obj
, "modern-hotplug-events",
2502 "Use dedicated hotplug event mechanism in"
2503 " place of standard EPOW events when possible"
2504 " (required for memory hot-unplug support)",
2508 static void spapr_machine_finalizefn(Object
*obj
)
2510 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
2512 g_free(spapr
->kvm_type
);
2515 void spapr_do_system_reset_on_cpu(CPUState
*cs
, run_on_cpu_data arg
)
2517 cpu_synchronize_state(cs
);
2518 ppc_cpu_do_system_reset(cs
);
2521 static void spapr_nmi(NMIState
*n
, int cpu_index
, Error
**errp
)
2526 async_run_on_cpu(cs
, spapr_do_system_reset_on_cpu
, RUN_ON_CPU_NULL
);
2530 static void spapr_add_lmbs(DeviceState
*dev
, uint64_t addr_start
, uint64_t size
,
2531 uint32_t node
, bool dedicated_hp_event_source
,
2534 sPAPRDRConnector
*drc
;
2535 uint32_t nr_lmbs
= size
/SPAPR_MEMORY_BLOCK_SIZE
;
2536 int i
, fdt_offset
, fdt_size
;
2538 uint64_t addr
= addr_start
;
2540 for (i
= 0; i
< nr_lmbs
; i
++) {
2541 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
2542 addr
/ SPAPR_MEMORY_BLOCK_SIZE
);
2545 fdt
= create_device_tree(&fdt_size
);
2546 fdt_offset
= spapr_populate_memory_node(fdt
, node
, addr
,
2547 SPAPR_MEMORY_BLOCK_SIZE
);
2549 spapr_drc_attach(drc
, dev
, fdt
, fdt_offset
, !dev
->hotplugged
, errp
);
2550 addr
+= SPAPR_MEMORY_BLOCK_SIZE
;
2551 if (!dev
->hotplugged
) {
2552 sPAPRDRConnectorClass
*drck
= SPAPR_DR_CONNECTOR_GET_CLASS(drc
);
2553 /* guests expect coldplugged LMBs to be pre-allocated */
2554 drck
->set_allocation_state(drc
, SPAPR_DR_ALLOCATION_STATE_USABLE
);
2555 drck
->set_isolation_state(drc
, SPAPR_DR_ISOLATION_STATE_UNISOLATED
);
2558 /* send hotplug notification to the
2559 * guest only in case of hotplugged memory
2561 if (dev
->hotplugged
) {
2562 if (dedicated_hp_event_source
) {
2563 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
2564 addr_start
/ SPAPR_MEMORY_BLOCK_SIZE
);
2565 spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB
,
2567 spapr_drc_index(drc
));
2569 spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB
,
2575 static void spapr_memory_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
2576 uint32_t node
, Error
**errp
)
2578 Error
*local_err
= NULL
;
2579 sPAPRMachineState
*ms
= SPAPR_MACHINE(hotplug_dev
);
2580 PCDIMMDevice
*dimm
= PC_DIMM(dev
);
2581 PCDIMMDeviceClass
*ddc
= PC_DIMM_GET_CLASS(dimm
);
2582 MemoryRegion
*mr
= ddc
->get_memory_region(dimm
);
2583 uint64_t align
= memory_region_get_alignment(mr
);
2584 uint64_t size
= memory_region_size(mr
);
2587 pc_dimm_memory_plug(dev
, &ms
->hotplug_memory
, mr
, align
, &local_err
);
2592 addr
= object_property_get_int(OBJECT(dimm
), PC_DIMM_ADDR_PROP
, &local_err
);
2594 pc_dimm_memory_unplug(dev
, &ms
->hotplug_memory
, mr
);
2598 spapr_add_lmbs(dev
, addr
, size
, node
,
2599 spapr_ovec_test(ms
->ov5_cas
, OV5_HP_EVT
),
2603 error_propagate(errp
, local_err
);
2606 static void spapr_memory_pre_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
2609 PCDIMMDevice
*dimm
= PC_DIMM(dev
);
2610 PCDIMMDeviceClass
*ddc
= PC_DIMM_GET_CLASS(dimm
);
2611 MemoryRegion
*mr
= ddc
->get_memory_region(dimm
);
2612 uint64_t size
= memory_region_size(mr
);
2615 if (size
% SPAPR_MEMORY_BLOCK_SIZE
) {
2616 error_setg(errp
, "Hotplugged memory size must be a multiple of "
2617 "%lld MB", SPAPR_MEMORY_BLOCK_SIZE
/ M_BYTE
);
2621 mem_dev
= object_property_get_str(OBJECT(dimm
), PC_DIMM_MEMDEV_PROP
, NULL
);
2622 if (mem_dev
&& !kvmppc_is_mem_backend_page_size_ok(mem_dev
)) {
2623 error_setg(errp
, "Memory backend has bad page size. "
2624 "Use 'memory-backend-file' with correct mem-path.");
2632 struct sPAPRDIMMState
{
2635 QTAILQ_ENTRY(sPAPRDIMMState
) next
;
2638 static sPAPRDIMMState
*spapr_pending_dimm_unplugs_find(sPAPRMachineState
*s
,
2641 sPAPRDIMMState
*dimm_state
= NULL
;
2643 QTAILQ_FOREACH(dimm_state
, &s
->pending_dimm_unplugs
, next
) {
2644 if (dimm_state
->dimm
== dimm
) {
2651 static void spapr_pending_dimm_unplugs_add(sPAPRMachineState
*spapr
,
2652 sPAPRDIMMState
*dimm_state
)
2654 g_assert(!spapr_pending_dimm_unplugs_find(spapr
, dimm_state
->dimm
));
2655 QTAILQ_INSERT_HEAD(&spapr
->pending_dimm_unplugs
, dimm_state
, next
);
2658 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState
*spapr
,
2659 sPAPRDIMMState
*dimm_state
)
2661 QTAILQ_REMOVE(&spapr
->pending_dimm_unplugs
, dimm_state
, next
);
2665 static sPAPRDIMMState
*spapr_recover_pending_dimm_state(sPAPRMachineState
*ms
,
2668 sPAPRDRConnector
*drc
;
2669 PCDIMMDeviceClass
*ddc
= PC_DIMM_GET_CLASS(dimm
);
2670 MemoryRegion
*mr
= ddc
->get_memory_region(dimm
);
2671 uint64_t size
= memory_region_size(mr
);
2672 uint32_t nr_lmbs
= size
/ SPAPR_MEMORY_BLOCK_SIZE
;
2673 uint32_t avail_lmbs
= 0;
2674 uint64_t addr_start
, addr
;
2678 addr_start
= object_property_get_int(OBJECT(dimm
), PC_DIMM_ADDR_PROP
,
2682 for (i
= 0; i
< nr_lmbs
; i
++) {
2683 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
2684 addr
/ SPAPR_MEMORY_BLOCK_SIZE
);
2689 addr
+= SPAPR_MEMORY_BLOCK_SIZE
;
2692 ds
= g_malloc0(sizeof(sPAPRDIMMState
));
2693 ds
->nr_lmbs
= avail_lmbs
;
2695 spapr_pending_dimm_unplugs_add(ms
, ds
);
2699 /* Callback to be called during DRC release. */
2700 void spapr_lmb_release(DeviceState
*dev
)
2702 HotplugHandler
*hotplug_ctrl
= qdev_get_hotplug_handler(dev
);
2703 sPAPRMachineState
*spapr
= SPAPR_MACHINE(hotplug_ctrl
);
2704 sPAPRDIMMState
*ds
= spapr_pending_dimm_unplugs_find(spapr
, PC_DIMM(dev
));
2706 /* This information will get lost if a migration occurs
2707 * during the unplug process. In this case recover it. */
2709 ds
= spapr_recover_pending_dimm_state(spapr
, PC_DIMM(dev
));
2710 /* The DRC being examined by the caller at least must be counted */
2711 g_assert(ds
->nr_lmbs
);
2714 if (--ds
->nr_lmbs
) {
2718 spapr_pending_dimm_unplugs_remove(spapr
, ds
);
2721 * Now that all the LMBs have been removed by the guest, call the
2722 * pc-dimm unplug handler to cleanup up the pc-dimm device.
2724 hotplug_handler_unplug(hotplug_ctrl
, dev
, &error_abort
);
2727 static void spapr_memory_unplug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
2730 sPAPRMachineState
*ms
= SPAPR_MACHINE(hotplug_dev
);
2731 PCDIMMDevice
*dimm
= PC_DIMM(dev
);
2732 PCDIMMDeviceClass
*ddc
= PC_DIMM_GET_CLASS(dimm
);
2733 MemoryRegion
*mr
= ddc
->get_memory_region(dimm
);
2735 pc_dimm_memory_unplug(dev
, &ms
->hotplug_memory
, mr
);
2736 object_unparent(OBJECT(dev
));
2739 static void spapr_memory_unplug_request(HotplugHandler
*hotplug_dev
,
2740 DeviceState
*dev
, Error
**errp
)
2742 sPAPRMachineState
*spapr
= SPAPR_MACHINE(hotplug_dev
);
2743 Error
*local_err
= NULL
;
2744 PCDIMMDevice
*dimm
= PC_DIMM(dev
);
2745 PCDIMMDeviceClass
*ddc
= PC_DIMM_GET_CLASS(dimm
);
2746 MemoryRegion
*mr
= ddc
->get_memory_region(dimm
);
2747 uint64_t size
= memory_region_size(mr
);
2748 uint32_t nr_lmbs
= size
/ SPAPR_MEMORY_BLOCK_SIZE
;
2749 uint64_t addr_start
, addr
;
2751 sPAPRDRConnector
*drc
;
2754 addr_start
= object_property_get_int(OBJECT(dimm
), PC_DIMM_ADDR_PROP
,
2760 ds
= g_malloc0(sizeof(sPAPRDIMMState
));
2761 ds
->nr_lmbs
= nr_lmbs
;
2763 spapr_pending_dimm_unplugs_add(spapr
, ds
);
2766 for (i
= 0; i
< nr_lmbs
; i
++) {
2767 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
2768 addr
/ SPAPR_MEMORY_BLOCK_SIZE
);
2771 spapr_drc_detach(drc
, dev
, errp
);
2772 addr
+= SPAPR_MEMORY_BLOCK_SIZE
;
2775 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_LMB
,
2776 addr_start
/ SPAPR_MEMORY_BLOCK_SIZE
);
2777 spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB
,
2778 nr_lmbs
, spapr_drc_index(drc
));
2780 error_propagate(errp
, local_err
);
2783 void *spapr_populate_hotplug_cpu_dt(CPUState
*cs
, int *fdt_offset
,
2784 sPAPRMachineState
*spapr
)
2786 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
2787 DeviceClass
*dc
= DEVICE_GET_CLASS(cs
);
2788 int id
= ppc_get_vcpu_dt_id(cpu
);
2790 int offset
, fdt_size
;
2793 fdt
= create_device_tree(&fdt_size
);
2794 nodename
= g_strdup_printf("%s@%x", dc
->fw_name
, id
);
2795 offset
= fdt_add_subnode(fdt
, 0, nodename
);
2797 spapr_populate_cpu_dt(cs
, fdt
, offset
, spapr
);
2800 *fdt_offset
= offset
;
2804 static void spapr_core_unplug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
2807 MachineState
*ms
= MACHINE(qdev_get_machine());
2808 CPUCore
*cc
= CPU_CORE(dev
);
2809 CPUArchId
*core_slot
= spapr_find_cpu_slot(ms
, cc
->core_id
, NULL
);
2812 core_slot
->cpu
= NULL
;
2813 object_unparent(OBJECT(dev
));
2816 /* Callback to be called during DRC release. */
2817 void spapr_core_release(DeviceState
*dev
)
2819 HotplugHandler
*hotplug_ctrl
;
2821 hotplug_ctrl
= qdev_get_hotplug_handler(dev
);
2822 hotplug_handler_unplug(hotplug_ctrl
, dev
, &error_abort
);
2826 void spapr_core_unplug_request(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
2830 sPAPRDRConnector
*drc
;
2831 Error
*local_err
= NULL
;
2832 CPUCore
*cc
= CPU_CORE(dev
);
2833 int smt
= kvmppc_smt_threads();
2835 if (!spapr_find_cpu_slot(MACHINE(hotplug_dev
), cc
->core_id
, &index
)) {
2836 error_setg(errp
, "Unable to find CPU core with core-id: %d",
2841 error_setg(errp
, "Boot CPU core may not be unplugged");
2845 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_CPU
, index
* smt
);
2848 spapr_drc_detach(drc
, dev
, &local_err
);
2850 error_propagate(errp
, local_err
);
2854 spapr_hotplug_req_remove_by_index(drc
);
2857 static void spapr_core_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
2860 sPAPRMachineState
*spapr
= SPAPR_MACHINE(OBJECT(hotplug_dev
));
2861 MachineClass
*mc
= MACHINE_GET_CLASS(spapr
);
2862 sPAPRCPUCore
*core
= SPAPR_CPU_CORE(OBJECT(dev
));
2863 CPUCore
*cc
= CPU_CORE(dev
);
2864 CPUState
*cs
= CPU(core
->threads
);
2865 sPAPRDRConnector
*drc
;
2866 Error
*local_err
= NULL
;
2869 int smt
= kvmppc_smt_threads();
2870 CPUArchId
*core_slot
;
2873 core_slot
= spapr_find_cpu_slot(MACHINE(hotplug_dev
), cc
->core_id
, &index
);
2875 error_setg(errp
, "Unable to find CPU core with core-id: %d",
2879 drc
= spapr_drc_by_id(TYPE_SPAPR_DRC_CPU
, index
* smt
);
2881 g_assert(drc
|| !mc
->has_hotpluggable_cpus
);
2884 * Setup CPU DT entries only for hotplugged CPUs. For boot time or
2885 * coldplugged CPUs DT entries are setup in spapr_build_fdt().
2887 if (dev
->hotplugged
) {
2888 fdt
= spapr_populate_hotplug_cpu_dt(cs
, &fdt_offset
, spapr
);
2892 spapr_drc_attach(drc
, dev
, fdt
, fdt_offset
, !dev
->hotplugged
,
2896 error_propagate(errp
, local_err
);
2901 if (dev
->hotplugged
) {
2903 * Send hotplug notification interrupt to the guest only in case
2904 * of hotplugged CPUs.
2906 spapr_hotplug_req_add_by_index(drc
);
2909 * Set the right DRC states for cold plugged CPU.
2912 sPAPRDRConnectorClass
*drck
= SPAPR_DR_CONNECTOR_GET_CLASS(drc
);
2913 drck
->set_allocation_state(drc
, SPAPR_DR_ALLOCATION_STATE_USABLE
);
2914 drck
->set_isolation_state(drc
, SPAPR_DR_ISOLATION_STATE_UNISOLATED
);
2917 core_slot
->cpu
= OBJECT(dev
);
2920 static void spapr_core_pre_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
2923 MachineState
*machine
= MACHINE(OBJECT(hotplug_dev
));
2924 MachineClass
*mc
= MACHINE_GET_CLASS(hotplug_dev
);
2925 Error
*local_err
= NULL
;
2926 CPUCore
*cc
= CPU_CORE(dev
);
2927 char *base_core_type
= spapr_get_cpu_core_type(machine
->cpu_model
);
2928 const char *type
= object_get_typename(OBJECT(dev
));
2929 CPUArchId
*core_slot
;
2932 if (dev
->hotplugged
&& !mc
->has_hotpluggable_cpus
) {
2933 error_setg(&local_err
, "CPU hotplug not supported for this machine");
2937 if (strcmp(base_core_type
, type
)) {
2938 error_setg(&local_err
, "CPU core type should be %s", base_core_type
);
2942 if (cc
->core_id
% smp_threads
) {
2943 error_setg(&local_err
, "invalid core id %d", cc
->core_id
);
2948 * In general we should have homogeneous threads-per-core, but old
2949 * (pre hotplug support) machine types allow the last core to have
2950 * reduced threads as a compatibility hack for when we allowed
2951 * total vcpus not a multiple of threads-per-core.
2953 if (mc
->has_hotpluggable_cpus
&& (cc
->nr_threads
!= smp_threads
)) {
2954 error_setg(errp
, "invalid nr-threads %d, must be %d",
2955 cc
->nr_threads
, smp_threads
);
2959 core_slot
= spapr_find_cpu_slot(MACHINE(hotplug_dev
), cc
->core_id
, &index
);
2961 error_setg(&local_err
, "core id %d out of range", cc
->core_id
);
2965 if (core_slot
->cpu
) {
2966 error_setg(&local_err
, "core %d already populated", cc
->core_id
);
2970 numa_cpu_pre_plug(core_slot
, dev
, &local_err
);
2973 g_free(base_core_type
);
2974 error_propagate(errp
, local_err
);
2977 static void spapr_machine_device_plug(HotplugHandler
*hotplug_dev
,
2978 DeviceState
*dev
, Error
**errp
)
2980 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2982 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
)) {
2985 if (!smc
->dr_lmb_enabled
) {
2986 error_setg(errp
, "Memory hotplug not supported for this machine");
2989 node
= object_property_get_int(OBJECT(dev
), PC_DIMM_NODE_PROP
, errp
);
2993 if (node
< 0 || node
>= MAX_NODES
) {
2994 error_setg(errp
, "Invaild node %d", node
);
2999 * Currently PowerPC kernel doesn't allow hot-adding memory to
3000 * memory-less node, but instead will silently add the memory
3001 * to the first node that has some memory. This causes two
3002 * unexpected behaviours for the user.
3004 * - Memory gets hotplugged to a different node than what the user
3006 * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
3007 * to memory-less node, a reboot will set things accordingly
3008 * and the previously hotplugged memory now ends in the right node.
3009 * This appears as if some memory moved from one node to another.
3011 * So until kernel starts supporting memory hotplug to memory-less
3012 * nodes, just prevent such attempts upfront in QEMU.
3014 if (nb_numa_nodes
&& !numa_info
[node
].node_mem
) {
3015 error_setg(errp
, "Can't hotplug memory to memory-less node %d",
3020 spapr_memory_plug(hotplug_dev
, dev
, node
, errp
);
3021 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
3022 spapr_core_plug(hotplug_dev
, dev
, errp
);
3026 static void spapr_machine_device_unplug(HotplugHandler
*hotplug_dev
,
3027 DeviceState
*dev
, Error
**errp
)
3029 sPAPRMachineState
*sms
= SPAPR_MACHINE(qdev_get_machine());
3030 MachineClass
*mc
= MACHINE_GET_CLASS(qdev_get_machine());
3032 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
)) {
3033 if (spapr_ovec_test(sms
->ov5_cas
, OV5_HP_EVT
)) {
3034 spapr_memory_unplug(hotplug_dev
, dev
, errp
);
3036 error_setg(errp
, "Memory hot unplug not supported for this guest");
3038 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
3039 if (!mc
->has_hotpluggable_cpus
) {
3040 error_setg(errp
, "CPU hot unplug not supported on this machine");
3043 spapr_core_unplug(hotplug_dev
, dev
, errp
);
3047 static void spapr_machine_device_unplug_request(HotplugHandler
*hotplug_dev
,
3048 DeviceState
*dev
, Error
**errp
)
3050 sPAPRMachineState
*sms
= SPAPR_MACHINE(qdev_get_machine());
3051 MachineClass
*mc
= MACHINE_GET_CLASS(qdev_get_machine());
3053 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
)) {
3054 if (spapr_ovec_test(sms
->ov5_cas
, OV5_HP_EVT
)) {
3055 spapr_memory_unplug_request(hotplug_dev
, dev
, errp
);
3057 /* NOTE: this means there is a window after guest reset, prior to
3058 * CAS negotiation, where unplug requests will fail due to the
3059 * capability not being detected yet. This is a bit different than
3060 * the case with PCI unplug, where the events will be queued and
3061 * eventually handled by the guest after boot
3063 error_setg(errp
, "Memory hot unplug not supported for this guest");
3065 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
3066 if (!mc
->has_hotpluggable_cpus
) {
3067 error_setg(errp
, "CPU hot unplug not supported on this machine");
3070 spapr_core_unplug_request(hotplug_dev
, dev
, errp
);
3074 static void spapr_machine_device_pre_plug(HotplugHandler
*hotplug_dev
,
3075 DeviceState
*dev
, Error
**errp
)
3077 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
)) {
3078 spapr_memory_pre_plug(hotplug_dev
, dev
, errp
);
3079 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
3080 spapr_core_pre_plug(hotplug_dev
, dev
, errp
);
3084 static HotplugHandler
*spapr_get_hotplug_handler(MachineState
*machine
,
3087 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
) ||
3088 object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
3089 return HOTPLUG_HANDLER(machine
);
3094 static CpuInstanceProperties
3095 spapr_cpu_index_to_props(MachineState
*machine
, unsigned cpu_index
)
3097 CPUArchId
*core_slot
;
3098 MachineClass
*mc
= MACHINE_GET_CLASS(machine
);
3100 /* make sure possible_cpu are intialized */
3101 mc
->possible_cpu_arch_ids(machine
);
3102 /* get CPU core slot containing thread that matches cpu_index */
3103 core_slot
= spapr_find_cpu_slot(machine
, cpu_index
, NULL
);
3105 return core_slot
->props
;
3108 static const CPUArchIdList
*spapr_possible_cpu_arch_ids(MachineState
*machine
)
3111 int spapr_max_cores
= max_cpus
/ smp_threads
;
3112 MachineClass
*mc
= MACHINE_GET_CLASS(machine
);
3114 if (!mc
->has_hotpluggable_cpus
) {
3115 spapr_max_cores
= QEMU_ALIGN_UP(smp_cpus
, smp_threads
) / smp_threads
;
3117 if (machine
->possible_cpus
) {
3118 assert(machine
->possible_cpus
->len
== spapr_max_cores
);
3119 return machine
->possible_cpus
;
3122 machine
->possible_cpus
= g_malloc0(sizeof(CPUArchIdList
) +
3123 sizeof(CPUArchId
) * spapr_max_cores
);
3124 machine
->possible_cpus
->len
= spapr_max_cores
;
3125 for (i
= 0; i
< machine
->possible_cpus
->len
; i
++) {
3126 int core_id
= i
* smp_threads
;
3128 machine
->possible_cpus
->cpus
[i
].vcpus_count
= smp_threads
;
3129 machine
->possible_cpus
->cpus
[i
].arch_id
= core_id
;
3130 machine
->possible_cpus
->cpus
[i
].props
.has_core_id
= true;
3131 machine
->possible_cpus
->cpus
[i
].props
.core_id
= core_id
;
3133 /* default distribution of CPUs over NUMA nodes */
3134 if (nb_numa_nodes
) {
3135 /* preset values but do not enable them i.e. 'has_node_id = false',
3136 * numa init code will enable them later if manual mapping wasn't
3138 machine
->possible_cpus
->cpus
[i
].props
.node_id
=
3139 core_id
/ smp_threads
/ smp_cores
% nb_numa_nodes
;
3142 return machine
->possible_cpus
;
3145 static void spapr_phb_placement(sPAPRMachineState
*spapr
, uint32_t index
,
3146 uint64_t *buid
, hwaddr
*pio
,
3147 hwaddr
*mmio32
, hwaddr
*mmio64
,
3148 unsigned n_dma
, uint32_t *liobns
, Error
**errp
)
3151 * New-style PHB window placement.
3153 * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3154 * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3157 * Some guest kernels can't work with MMIO windows above 1<<46
3158 * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3160 * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3161 * PHB stacked together. (32TiB+2GiB)..(32TiB+64GiB) contains the
3162 * 2GiB 32-bit MMIO windows for each PHB. Then 33..64TiB has the
3163 * 1TiB 64-bit MMIO windows for each PHB.
3165 const uint64_t base_buid
= 0x800000020000000ULL
;
3166 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3167 SPAPR_PCI_MEM64_WIN_SIZE - 1)
3170 /* Sanity check natural alignments */
3171 QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE
% SPAPR_PCI_MEM64_WIN_SIZE
) != 0);
3172 QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT
% SPAPR_PCI_MEM64_WIN_SIZE
) != 0);
3173 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE
% SPAPR_PCI_MEM32_WIN_SIZE
) != 0);
3174 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE
% SPAPR_PCI_IO_WIN_SIZE
) != 0);
3175 /* Sanity check bounds */
3176 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS
* SPAPR_PCI_IO_WIN_SIZE
) >
3177 SPAPR_PCI_MEM32_WIN_SIZE
);
3178 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS
* SPAPR_PCI_MEM32_WIN_SIZE
) >
3179 SPAPR_PCI_MEM64_WIN_SIZE
);
3181 if (index
>= SPAPR_MAX_PHBS
) {
3182 error_setg(errp
, "\"index\" for PAPR PHB is too large (max %llu)",
3183 SPAPR_MAX_PHBS
- 1);
3187 *buid
= base_buid
+ index
;
3188 for (i
= 0; i
< n_dma
; ++i
) {
3189 liobns
[i
] = SPAPR_PCI_LIOBN(index
, i
);
3192 *pio
= SPAPR_PCI_BASE
+ index
* SPAPR_PCI_IO_WIN_SIZE
;
3193 *mmio32
= SPAPR_PCI_BASE
+ (index
+ 1) * SPAPR_PCI_MEM32_WIN_SIZE
;
3194 *mmio64
= SPAPR_PCI_BASE
+ (index
+ 1) * SPAPR_PCI_MEM64_WIN_SIZE
;
3197 static ICSState
*spapr_ics_get(XICSFabric
*dev
, int irq
)
3199 sPAPRMachineState
*spapr
= SPAPR_MACHINE(dev
);
3201 return ics_valid_irq(spapr
->ics
, irq
) ? spapr
->ics
: NULL
;
3204 static void spapr_ics_resend(XICSFabric
*dev
)
3206 sPAPRMachineState
*spapr
= SPAPR_MACHINE(dev
);
3208 ics_resend(spapr
->ics
);
3211 static ICPState
*spapr_icp_get(XICSFabric
*xi
, int cpu_dt_id
)
3213 PowerPCCPU
*cpu
= ppc_get_vcpu_by_dt_id(cpu_dt_id
);
3215 return cpu
? ICP(cpu
->intc
) : NULL
;
3218 static void spapr_pic_print_info(InterruptStatsProvider
*obj
,
3221 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
3225 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
3227 icp_pic_print_info(ICP(cpu
->intc
), mon
);
3230 ics_pic_print_info(spapr
->ics
, mon
);
3233 static void spapr_machine_class_init(ObjectClass
*oc
, void *data
)
3235 MachineClass
*mc
= MACHINE_CLASS(oc
);
3236 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(oc
);
3237 FWPathProviderClass
*fwc
= FW_PATH_PROVIDER_CLASS(oc
);
3238 NMIClass
*nc
= NMI_CLASS(oc
);
3239 HotplugHandlerClass
*hc
= HOTPLUG_HANDLER_CLASS(oc
);
3240 PPCVirtualHypervisorClass
*vhc
= PPC_VIRTUAL_HYPERVISOR_CLASS(oc
);
3241 XICSFabricClass
*xic
= XICS_FABRIC_CLASS(oc
);
3242 InterruptStatsProviderClass
*ispc
= INTERRUPT_STATS_PROVIDER_CLASS(oc
);
3244 mc
->desc
= "pSeries Logical Partition (PAPR compliant)";
3247 * We set up the default / latest behaviour here. The class_init
3248 * functions for the specific versioned machine types can override
3249 * these details for backwards compatibility
3251 mc
->init
= ppc_spapr_init
;
3252 mc
->reset
= ppc_spapr_reset
;
3253 mc
->block_default_type
= IF_SCSI
;
3254 mc
->max_cpus
= 1024;
3255 mc
->no_parallel
= 1;
3256 mc
->default_boot_order
= "";
3257 mc
->default_ram_size
= 512 * M_BYTE
;
3258 mc
->kvm_type
= spapr_kvm_type
;
3259 mc
->has_dynamic_sysbus
= true;
3260 mc
->pci_allow_0_address
= true;
3261 mc
->get_hotplug_handler
= spapr_get_hotplug_handler
;
3262 hc
->pre_plug
= spapr_machine_device_pre_plug
;
3263 hc
->plug
= spapr_machine_device_plug
;
3264 hc
->unplug
= spapr_machine_device_unplug
;
3265 mc
->cpu_index_to_instance_props
= spapr_cpu_index_to_props
;
3266 mc
->possible_cpu_arch_ids
= spapr_possible_cpu_arch_ids
;
3267 hc
->unplug_request
= spapr_machine_device_unplug_request
;
3269 smc
->dr_lmb_enabled
= true;
3270 smc
->tcg_default_cpu
= "POWER8";
3271 mc
->has_hotpluggable_cpus
= true;
3272 fwc
->get_dev_path
= spapr_get_fw_dev_path
;
3273 nc
->nmi_monitor_handler
= spapr_nmi
;
3274 smc
->phb_placement
= spapr_phb_placement
;
3275 vhc
->hypercall
= emulate_spapr_hypercall
;
3276 vhc
->hpt_mask
= spapr_hpt_mask
;
3277 vhc
->map_hptes
= spapr_map_hptes
;
3278 vhc
->unmap_hptes
= spapr_unmap_hptes
;
3279 vhc
->store_hpte
= spapr_store_hpte
;
3280 vhc
->get_patbe
= spapr_get_patbe
;
3281 xic
->ics_get
= spapr_ics_get
;
3282 xic
->ics_resend
= spapr_ics_resend
;
3283 xic
->icp_get
= spapr_icp_get
;
3284 ispc
->print_info
= spapr_pic_print_info
;
3285 /* Force NUMA node memory size to be a multiple of
3286 * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3287 * in which LMBs are represented and hot-added
3289 mc
->numa_mem_align_shift
= 28;
3292 static const TypeInfo spapr_machine_info
= {
3293 .name
= TYPE_SPAPR_MACHINE
,
3294 .parent
= TYPE_MACHINE
,
3296 .instance_size
= sizeof(sPAPRMachineState
),
3297 .instance_init
= spapr_machine_initfn
,
3298 .instance_finalize
= spapr_machine_finalizefn
,
3299 .class_size
= sizeof(sPAPRMachineClass
),
3300 .class_init
= spapr_machine_class_init
,
3301 .interfaces
= (InterfaceInfo
[]) {
3302 { TYPE_FW_PATH_PROVIDER
},
3304 { TYPE_HOTPLUG_HANDLER
},
3305 { TYPE_PPC_VIRTUAL_HYPERVISOR
},
3306 { TYPE_XICS_FABRIC
},
3307 { TYPE_INTERRUPT_STATS_PROVIDER
},
3312 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest) \
3313 static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3316 MachineClass *mc = MACHINE_CLASS(oc); \
3317 spapr_machine_##suffix##_class_options(mc); \
3319 mc->alias = "pseries"; \
3320 mc->is_default = 1; \
3323 static void spapr_machine_##suffix##_instance_init(Object *obj) \
3325 MachineState *machine = MACHINE(obj); \
3326 spapr_machine_##suffix##_instance_options(machine); \
3328 static const TypeInfo spapr_machine_##suffix##_info = { \
3329 .name = MACHINE_TYPE_NAME("pseries-" verstr), \
3330 .parent = TYPE_SPAPR_MACHINE, \
3331 .class_init = spapr_machine_##suffix##_class_init, \
3332 .instance_init = spapr_machine_##suffix##_instance_init, \
3334 static void spapr_machine_register_##suffix(void) \
3336 type_register(&spapr_machine_##suffix##_info); \
3338 type_init(spapr_machine_register_##suffix)
3343 static void spapr_machine_2_10_instance_options(MachineState
*machine
)
3347 static void spapr_machine_2_10_class_options(MachineClass
*mc
)
3349 /* Defaults for the latest behaviour inherited from the base class */
3352 DEFINE_SPAPR_MACHINE(2_10
, "2.10", true);
3357 #define SPAPR_COMPAT_2_9 \
3360 static void spapr_machine_2_9_instance_options(MachineState
*machine
)
3362 spapr_machine_2_10_instance_options(machine
);
3365 static void spapr_machine_2_9_class_options(MachineClass
*mc
)
3367 spapr_machine_2_10_class_options(mc
);
3368 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_9
);
3369 mc
->numa_auto_assign_ram
= numa_legacy_auto_assign_ram
;
3372 DEFINE_SPAPR_MACHINE(2_9
, "2.9", false);
3377 #define SPAPR_COMPAT_2_8 \
3380 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
3381 .property = "pcie-extended-configuration-space", \
3385 static void spapr_machine_2_8_instance_options(MachineState
*machine
)
3387 spapr_machine_2_9_instance_options(machine
);
3390 static void spapr_machine_2_8_class_options(MachineClass
*mc
)
3392 spapr_machine_2_9_class_options(mc
);
3393 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_8
);
3394 mc
->numa_mem_align_shift
= 23;
3397 DEFINE_SPAPR_MACHINE(2_8
, "2.8", false);
3402 #define SPAPR_COMPAT_2_7 \
3405 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
3406 .property = "mem_win_size", \
3407 .value = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3410 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
3411 .property = "mem64_win_size", \
3415 .driver = TYPE_POWERPC_CPU, \
3416 .property = "pre-2.8-migration", \
3420 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
3421 .property = "pre-2.8-migration", \
3425 static void phb_placement_2_7(sPAPRMachineState
*spapr
, uint32_t index
,
3426 uint64_t *buid
, hwaddr
*pio
,
3427 hwaddr
*mmio32
, hwaddr
*mmio64
,
3428 unsigned n_dma
, uint32_t *liobns
, Error
**errp
)
3430 /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3431 const uint64_t base_buid
= 0x800000020000000ULL
;
3432 const hwaddr phb_spacing
= 0x1000000000ULL
; /* 64 GiB */
3433 const hwaddr mmio_offset
= 0xa0000000; /* 2 GiB + 512 MiB */
3434 const hwaddr pio_offset
= 0x80000000; /* 2 GiB */
3435 const uint32_t max_index
= 255;
3436 const hwaddr phb0_alignment
= 0x10000000000ULL
; /* 1 TiB */
3438 uint64_t ram_top
= MACHINE(spapr
)->ram_size
;
3439 hwaddr phb0_base
, phb_base
;
3442 /* Do we have hotpluggable memory? */
3443 if (MACHINE(spapr
)->maxram_size
> ram_top
) {
3444 /* Can't just use maxram_size, because there may be an
3445 * alignment gap between normal and hotpluggable memory
3447 ram_top
= spapr
->hotplug_memory
.base
+
3448 memory_region_size(&spapr
->hotplug_memory
.mr
);
3451 phb0_base
= QEMU_ALIGN_UP(ram_top
, phb0_alignment
);
3453 if (index
> max_index
) {
3454 error_setg(errp
, "\"index\" for PAPR PHB is too large (max %u)",
3459 *buid
= base_buid
+ index
;
3460 for (i
= 0; i
< n_dma
; ++i
) {
3461 liobns
[i
] = SPAPR_PCI_LIOBN(index
, i
);
3464 phb_base
= phb0_base
+ index
* phb_spacing
;
3465 *pio
= phb_base
+ pio_offset
;
3466 *mmio32
= phb_base
+ mmio_offset
;
3468 * We don't set the 64-bit MMIO window, relying on the PHB's
3469 * fallback behaviour of automatically splitting a large "32-bit"
3470 * window into contiguous 32-bit and 64-bit windows
3474 static void spapr_machine_2_7_instance_options(MachineState
*machine
)
3476 sPAPRMachineState
*spapr
= SPAPR_MACHINE(machine
);
3478 spapr_machine_2_8_instance_options(machine
);
3479 spapr
->use_hotplug_event_source
= false;
3482 static void spapr_machine_2_7_class_options(MachineClass
*mc
)
3484 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
3486 spapr_machine_2_8_class_options(mc
);
3487 smc
->tcg_default_cpu
= "POWER7";
3488 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_7
);
3489 smc
->phb_placement
= phb_placement_2_7
;
3492 DEFINE_SPAPR_MACHINE(2_7
, "2.7", false);
3497 #define SPAPR_COMPAT_2_6 \
3500 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3502 .value = stringify(off),\
3505 static void spapr_machine_2_6_instance_options(MachineState
*machine
)
3507 spapr_machine_2_7_instance_options(machine
);
3510 static void spapr_machine_2_6_class_options(MachineClass
*mc
)
3512 spapr_machine_2_7_class_options(mc
);
3513 mc
->has_hotpluggable_cpus
= false;
3514 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_6
);
3517 DEFINE_SPAPR_MACHINE(2_6
, "2.6", false);
3522 #define SPAPR_COMPAT_2_5 \
3525 .driver = "spapr-vlan", \
3526 .property = "use-rx-buffer-pools", \
3530 static void spapr_machine_2_5_instance_options(MachineState
*machine
)
3532 spapr_machine_2_6_instance_options(machine
);
3535 static void spapr_machine_2_5_class_options(MachineClass
*mc
)
3537 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
3539 spapr_machine_2_6_class_options(mc
);
3540 smc
->use_ohci_by_default
= true;
3541 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_5
);
3544 DEFINE_SPAPR_MACHINE(2_5
, "2.5", false);
3549 #define SPAPR_COMPAT_2_4 \
3552 static void spapr_machine_2_4_instance_options(MachineState
*machine
)
3554 spapr_machine_2_5_instance_options(machine
);
3557 static void spapr_machine_2_4_class_options(MachineClass
*mc
)
3559 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
3561 spapr_machine_2_5_class_options(mc
);
3562 smc
->dr_lmb_enabled
= false;
3563 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_4
);
3566 DEFINE_SPAPR_MACHINE(2_4
, "2.4", false);
3571 #define SPAPR_COMPAT_2_3 \
3574 .driver = "spapr-pci-host-bridge",\
3575 .property = "dynamic-reconfiguration",\
3579 static void spapr_machine_2_3_instance_options(MachineState
*machine
)
3581 spapr_machine_2_4_instance_options(machine
);
3582 savevm_skip_section_footers();
3583 global_state_set_optional();
3584 savevm_skip_configuration();
3587 static void spapr_machine_2_3_class_options(MachineClass
*mc
)
3589 spapr_machine_2_4_class_options(mc
);
3590 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_3
);
3592 DEFINE_SPAPR_MACHINE(2_3
, "2.3", false);
3598 #define SPAPR_COMPAT_2_2 \
3601 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3602 .property = "mem_win_size",\
3603 .value = "0x20000000",\
3606 static void spapr_machine_2_2_instance_options(MachineState
*machine
)
3608 spapr_machine_2_3_instance_options(machine
);
3609 machine
->suppress_vmdesc
= true;
3612 static void spapr_machine_2_2_class_options(MachineClass
*mc
)
3614 spapr_machine_2_3_class_options(mc
);
3615 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_2
);
3617 DEFINE_SPAPR_MACHINE(2_2
, "2.2", false);
3622 #define SPAPR_COMPAT_2_1 \
3625 static void spapr_machine_2_1_instance_options(MachineState
*machine
)
3627 spapr_machine_2_2_instance_options(machine
);
3630 static void spapr_machine_2_1_class_options(MachineClass
*mc
)
3632 spapr_machine_2_2_class_options(mc
);
3633 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_1
);
3635 DEFINE_SPAPR_MACHINE(2_1
, "2.1", false);
3637 static void spapr_machine_register_types(void)
3639 type_register_static(&spapr_machine_info
);
3642 type_init(spapr_machine_register_types
)