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/kvm.h"
40 #include "sysemu/device_tree.h"
42 #include "migration/migration.h"
43 #include "mmu-hash64.h"
46 #include "hw/boards.h"
47 #include "hw/ppc/ppc.h"
48 #include "hw/loader.h"
50 #include "hw/ppc/spapr.h"
51 #include "hw/ppc/spapr_vio.h"
52 #include "hw/pci-host/spapr.h"
53 #include "hw/ppc/xics.h"
54 #include "hw/pci/msi.h"
56 #include "hw/pci/pci.h"
57 #include "hw/scsi/scsi.h"
58 #include "hw/virtio/virtio-scsi.h"
60 #include "exec/address-spaces.h"
62 #include "qemu/config-file.h"
63 #include "qemu/error-report.h"
67 #include "hw/compat.h"
68 #include "qemu/cutils.h"
69 #include "hw/ppc/spapr_cpu_core.h"
70 #include "qmp-commands.h"
74 /* SLOF memory layout:
76 * SLOF raw image loaded at 0, copies its romfs right below the flat
77 * device-tree, then position SLOF itself 31M below that
79 * So we set FW_OVERHEAD to 40MB which should account for all of that
82 * We load our kernel at 4M, leaving space for SLOF initial image
84 #define FDT_MAX_SIZE 0x100000
85 #define RTAS_MAX_SIZE 0x10000
86 #define RTAS_MAX_ADDR 0x80000000 /* RTAS must stay below that */
87 #define FW_MAX_SIZE 0x400000
88 #define FW_FILE_NAME "slof.bin"
89 #define FW_OVERHEAD 0x2800000
90 #define KERNEL_LOAD_ADDR FW_MAX_SIZE
92 #define MIN_RMA_SLOF 128UL
94 #define PHANDLE_XICP 0x00001111
96 #define HTAB_SIZE(spapr) (1ULL << ((spapr)->htab_shift))
98 static XICSState
*try_create_xics(const char *type
, int nr_servers
,
99 int nr_irqs
, Error
**errp
)
104 dev
= qdev_create(NULL
, type
);
105 qdev_prop_set_uint32(dev
, "nr_servers", nr_servers
);
106 qdev_prop_set_uint32(dev
, "nr_irqs", nr_irqs
);
107 object_property_set_bool(OBJECT(dev
), true, "realized", &err
);
109 error_propagate(errp
, err
);
110 object_unparent(OBJECT(dev
));
113 return XICS_COMMON(dev
);
116 static XICSState
*xics_system_init(MachineState
*machine
,
117 int nr_servers
, int nr_irqs
, Error
**errp
)
119 XICSState
*xics
= NULL
;
124 if (machine_kernel_irqchip_allowed(machine
)) {
125 xics
= try_create_xics(TYPE_XICS_SPAPR_KVM
, nr_servers
, nr_irqs
,
128 if (machine_kernel_irqchip_required(machine
) && !xics
) {
129 error_reportf_err(err
,
130 "kernel_irqchip requested but unavailable: ");
137 xics
= try_create_xics(TYPE_XICS_SPAPR
, nr_servers
, nr_irqs
, errp
);
143 static int spapr_fixup_cpu_smt_dt(void *fdt
, int offset
, PowerPCCPU
*cpu
,
147 uint32_t servers_prop
[smt_threads
];
148 uint32_t gservers_prop
[smt_threads
* 2];
149 int index
= ppc_get_vcpu_dt_id(cpu
);
151 if (cpu
->cpu_version
) {
152 ret
= fdt_setprop_cell(fdt
, offset
, "cpu-version", cpu
->cpu_version
);
158 /* Build interrupt servers and gservers properties */
159 for (i
= 0; i
< smt_threads
; i
++) {
160 servers_prop
[i
] = cpu_to_be32(index
+ i
);
161 /* Hack, direct the group queues back to cpu 0 */
162 gservers_prop
[i
*2] = cpu_to_be32(index
+ i
);
163 gservers_prop
[i
*2 + 1] = 0;
165 ret
= fdt_setprop(fdt
, offset
, "ibm,ppc-interrupt-server#s",
166 servers_prop
, sizeof(servers_prop
));
170 ret
= fdt_setprop(fdt
, offset
, "ibm,ppc-interrupt-gserver#s",
171 gservers_prop
, sizeof(gservers_prop
));
176 static int spapr_fixup_cpu_numa_dt(void *fdt
, int offset
, CPUState
*cs
)
179 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
180 int index
= ppc_get_vcpu_dt_id(cpu
);
181 uint32_t associativity
[] = {cpu_to_be32(0x5),
185 cpu_to_be32(cs
->numa_node
),
188 /* Advertise NUMA via ibm,associativity */
189 if (nb_numa_nodes
> 1) {
190 ret
= fdt_setprop(fdt
, offset
, "ibm,associativity", associativity
,
191 sizeof(associativity
));
197 static int spapr_fixup_cpu_dt(void *fdt
, sPAPRMachineState
*spapr
)
199 int ret
= 0, offset
, cpus_offset
;
202 int smt
= kvmppc_smt_threads();
203 uint32_t pft_size_prop
[] = {0, cpu_to_be32(spapr
->htab_shift
)};
206 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
207 DeviceClass
*dc
= DEVICE_GET_CLASS(cs
);
208 int index
= ppc_get_vcpu_dt_id(cpu
);
210 if ((index
% smt
) != 0) {
214 snprintf(cpu_model
, 32, "%s@%x", dc
->fw_name
, index
);
216 cpus_offset
= fdt_path_offset(fdt
, "/cpus");
217 if (cpus_offset
< 0) {
218 cpus_offset
= fdt_add_subnode(fdt
, fdt_path_offset(fdt
, "/"),
220 if (cpus_offset
< 0) {
224 offset
= fdt_subnode_offset(fdt
, cpus_offset
, cpu_model
);
226 offset
= fdt_add_subnode(fdt
, cpus_offset
, cpu_model
);
232 ret
= fdt_setprop(fdt
, offset
, "ibm,pft-size",
233 pft_size_prop
, sizeof(pft_size_prop
));
238 ret
= spapr_fixup_cpu_numa_dt(fdt
, offset
, cs
);
243 ret
= spapr_fixup_cpu_smt_dt(fdt
, offset
, cpu
,
244 ppc_get_compat_smt_threads(cpu
));
253 static size_t create_page_sizes_prop(CPUPPCState
*env
, uint32_t *prop
,
256 size_t maxcells
= maxsize
/ sizeof(uint32_t);
260 for (i
= 0; i
< PPC_PAGE_SIZES_MAX_SZ
; i
++) {
261 struct ppc_one_seg_page_size
*sps
= &env
->sps
.sps
[i
];
263 if (!sps
->page_shift
) {
266 for (count
= 0; count
< PPC_PAGE_SIZES_MAX_SZ
; count
++) {
267 if (sps
->enc
[count
].page_shift
== 0) {
271 if ((p
- prop
) >= (maxcells
- 3 - count
* 2)) {
274 *(p
++) = cpu_to_be32(sps
->page_shift
);
275 *(p
++) = cpu_to_be32(sps
->slb_enc
);
276 *(p
++) = cpu_to_be32(count
);
277 for (j
= 0; j
< count
; j
++) {
278 *(p
++) = cpu_to_be32(sps
->enc
[j
].page_shift
);
279 *(p
++) = cpu_to_be32(sps
->enc
[j
].pte_enc
);
283 return (p
- prop
) * sizeof(uint32_t);
286 static hwaddr
spapr_node0_size(void)
288 MachineState
*machine
= MACHINE(qdev_get_machine());
292 for (i
= 0; i
< nb_numa_nodes
; ++i
) {
293 if (numa_info
[i
].node_mem
) {
294 return MIN(pow2floor(numa_info
[i
].node_mem
),
299 return machine
->ram_size
;
306 fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
307 #exp, fdt_strerror(ret)); \
312 static void add_str(GString
*s
, const gchar
*s1
)
314 g_string_append_len(s
, s1
, strlen(s1
) + 1);
317 static void *spapr_create_fdt_skel(hwaddr initrd_base
,
321 const char *kernel_cmdline
,
325 uint32_t start_prop
= cpu_to_be32(initrd_base
);
326 uint32_t end_prop
= cpu_to_be32(initrd_base
+ initrd_size
);
327 GString
*hypertas
= g_string_sized_new(256);
328 GString
*qemu_hypertas
= g_string_sized_new(256);
329 uint32_t refpoints
[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
330 uint32_t interrupt_server_ranges_prop
[] = {0, cpu_to_be32(max_cpus
)};
331 unsigned char vec5
[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
334 add_str(hypertas
, "hcall-pft");
335 add_str(hypertas
, "hcall-term");
336 add_str(hypertas
, "hcall-dabr");
337 add_str(hypertas
, "hcall-interrupt");
338 add_str(hypertas
, "hcall-tce");
339 add_str(hypertas
, "hcall-vio");
340 add_str(hypertas
, "hcall-splpar");
341 add_str(hypertas
, "hcall-bulk");
342 add_str(hypertas
, "hcall-set-mode");
343 add_str(hypertas
, "hcall-sprg0");
344 add_str(hypertas
, "hcall-copy");
345 add_str(hypertas
, "hcall-debug");
346 add_str(qemu_hypertas
, "hcall-memop1");
348 fdt
= g_malloc0(FDT_MAX_SIZE
);
349 _FDT((fdt_create(fdt
, FDT_MAX_SIZE
)));
352 _FDT((fdt_add_reservemap_entry(fdt
, KERNEL_LOAD_ADDR
, kernel_size
)));
355 _FDT((fdt_add_reservemap_entry(fdt
, initrd_base
, initrd_size
)));
357 _FDT((fdt_finish_reservemap(fdt
)));
360 _FDT((fdt_begin_node(fdt
, "")));
361 _FDT((fdt_property_string(fdt
, "device_type", "chrp")));
362 _FDT((fdt_property_string(fdt
, "model", "IBM pSeries (emulated by qemu)")));
363 _FDT((fdt_property_string(fdt
, "compatible", "qemu,pseries")));
366 * Add info to guest to indentify which host is it being run on
367 * and what is the uuid of the guest
369 if (kvmppc_get_host_model(&buf
)) {
370 _FDT((fdt_property_string(fdt
, "host-model", buf
)));
373 if (kvmppc_get_host_serial(&buf
)) {
374 _FDT((fdt_property_string(fdt
, "host-serial", buf
)));
378 buf
= g_strdup_printf(UUID_FMT
, qemu_uuid
[0], qemu_uuid
[1],
379 qemu_uuid
[2], qemu_uuid
[3], qemu_uuid
[4],
380 qemu_uuid
[5], qemu_uuid
[6], qemu_uuid
[7],
381 qemu_uuid
[8], qemu_uuid
[9], qemu_uuid
[10],
382 qemu_uuid
[11], qemu_uuid
[12], qemu_uuid
[13],
383 qemu_uuid
[14], qemu_uuid
[15]);
385 _FDT((fdt_property_string(fdt
, "vm,uuid", buf
)));
387 _FDT((fdt_property_string(fdt
, "system-id", buf
)));
391 if (qemu_get_vm_name()) {
392 _FDT((fdt_property_string(fdt
, "ibm,partition-name",
393 qemu_get_vm_name())));
396 _FDT((fdt_property_cell(fdt
, "#address-cells", 0x2)));
397 _FDT((fdt_property_cell(fdt
, "#size-cells", 0x2)));
400 _FDT((fdt_begin_node(fdt
, "chosen")));
402 /* Set Form1_affinity */
403 _FDT((fdt_property(fdt
, "ibm,architecture-vec-5", vec5
, sizeof(vec5
))));
405 _FDT((fdt_property_string(fdt
, "bootargs", kernel_cmdline
)));
406 _FDT((fdt_property(fdt
, "linux,initrd-start",
407 &start_prop
, sizeof(start_prop
))));
408 _FDT((fdt_property(fdt
, "linux,initrd-end",
409 &end_prop
, sizeof(end_prop
))));
411 uint64_t kprop
[2] = { cpu_to_be64(KERNEL_LOAD_ADDR
),
412 cpu_to_be64(kernel_size
) };
414 _FDT((fdt_property(fdt
, "qemu,boot-kernel", &kprop
, sizeof(kprop
))));
416 _FDT((fdt_property(fdt
, "qemu,boot-kernel-le", NULL
, 0)));
420 _FDT((fdt_property_cell(fdt
, "qemu,boot-menu", boot_menu
)));
422 _FDT((fdt_property_cell(fdt
, "qemu,graphic-width", graphic_width
)));
423 _FDT((fdt_property_cell(fdt
, "qemu,graphic-height", graphic_height
)));
424 _FDT((fdt_property_cell(fdt
, "qemu,graphic-depth", graphic_depth
)));
426 _FDT((fdt_end_node(fdt
)));
429 _FDT((fdt_begin_node(fdt
, "rtas")));
431 if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
432 add_str(hypertas
, "hcall-multi-tce");
434 _FDT((fdt_property(fdt
, "ibm,hypertas-functions", hypertas
->str
,
436 g_string_free(hypertas
, TRUE
);
437 _FDT((fdt_property(fdt
, "qemu,hypertas-functions", qemu_hypertas
->str
,
438 qemu_hypertas
->len
)));
439 g_string_free(qemu_hypertas
, TRUE
);
441 _FDT((fdt_property(fdt
, "ibm,associativity-reference-points",
442 refpoints
, sizeof(refpoints
))));
444 _FDT((fdt_property_cell(fdt
, "rtas-error-log-max", RTAS_ERROR_LOG_MAX
)));
445 _FDT((fdt_property_cell(fdt
, "rtas-event-scan-rate",
446 RTAS_EVENT_SCAN_RATE
)));
449 _FDT((fdt_property(fdt
, "ibm,change-msix-capable", NULL
, 0)));
453 * According to PAPR, rtas ibm,os-term does not guarantee a return
454 * back to the guest cpu.
456 * While an additional ibm,extended-os-term property indicates that
457 * rtas call return will always occur. Set this property.
459 _FDT((fdt_property(fdt
, "ibm,extended-os-term", NULL
, 0)));
461 _FDT((fdt_end_node(fdt
)));
463 /* interrupt controller */
464 _FDT((fdt_begin_node(fdt
, "interrupt-controller")));
466 _FDT((fdt_property_string(fdt
, "device_type",
467 "PowerPC-External-Interrupt-Presentation")));
468 _FDT((fdt_property_string(fdt
, "compatible", "IBM,ppc-xicp")));
469 _FDT((fdt_property(fdt
, "interrupt-controller", NULL
, 0)));
470 _FDT((fdt_property(fdt
, "ibm,interrupt-server-ranges",
471 interrupt_server_ranges_prop
,
472 sizeof(interrupt_server_ranges_prop
))));
473 _FDT((fdt_property_cell(fdt
, "#interrupt-cells", 2)));
474 _FDT((fdt_property_cell(fdt
, "linux,phandle", PHANDLE_XICP
)));
475 _FDT((fdt_property_cell(fdt
, "phandle", PHANDLE_XICP
)));
477 _FDT((fdt_end_node(fdt
)));
480 _FDT((fdt_begin_node(fdt
, "vdevice")));
482 _FDT((fdt_property_string(fdt
, "device_type", "vdevice")));
483 _FDT((fdt_property_string(fdt
, "compatible", "IBM,vdevice")));
484 _FDT((fdt_property_cell(fdt
, "#address-cells", 0x1)));
485 _FDT((fdt_property_cell(fdt
, "#size-cells", 0x0)));
486 _FDT((fdt_property_cell(fdt
, "#interrupt-cells", 0x2)));
487 _FDT((fdt_property(fdt
, "interrupt-controller", NULL
, 0)));
489 _FDT((fdt_end_node(fdt
)));
492 spapr_events_fdt_skel(fdt
, epow_irq
);
494 /* /hypervisor node */
496 uint8_t hypercall
[16];
498 /* indicate KVM hypercall interface */
499 _FDT((fdt_begin_node(fdt
, "hypervisor")));
500 _FDT((fdt_property_string(fdt
, "compatible", "linux,kvm")));
501 if (kvmppc_has_cap_fixup_hcalls()) {
503 * Older KVM versions with older guest kernels were broken with the
504 * magic page, don't allow the guest to map it.
506 if (!kvmppc_get_hypercall(first_cpu
->env_ptr
, hypercall
,
507 sizeof(hypercall
))) {
508 _FDT((fdt_property(fdt
, "hcall-instructions", hypercall
,
509 sizeof(hypercall
))));
512 _FDT((fdt_end_node(fdt
)));
515 _FDT((fdt_end_node(fdt
))); /* close root node */
516 _FDT((fdt_finish(fdt
)));
521 static int spapr_populate_memory_node(void *fdt
, int nodeid
, hwaddr start
,
524 uint32_t associativity
[] = {
525 cpu_to_be32(0x4), /* length */
526 cpu_to_be32(0x0), cpu_to_be32(0x0),
527 cpu_to_be32(0x0), cpu_to_be32(nodeid
)
530 uint64_t mem_reg_property
[2];
533 mem_reg_property
[0] = cpu_to_be64(start
);
534 mem_reg_property
[1] = cpu_to_be64(size
);
536 sprintf(mem_name
, "memory@" TARGET_FMT_lx
, start
);
537 off
= fdt_add_subnode(fdt
, 0, mem_name
);
539 _FDT((fdt_setprop_string(fdt
, off
, "device_type", "memory")));
540 _FDT((fdt_setprop(fdt
, off
, "reg", mem_reg_property
,
541 sizeof(mem_reg_property
))));
542 _FDT((fdt_setprop(fdt
, off
, "ibm,associativity", associativity
,
543 sizeof(associativity
))));
547 static int spapr_populate_memory(sPAPRMachineState
*spapr
, void *fdt
)
549 MachineState
*machine
= MACHINE(spapr
);
550 hwaddr mem_start
, node_size
;
551 int i
, nb_nodes
= nb_numa_nodes
;
552 NodeInfo
*nodes
= numa_info
;
555 /* No NUMA nodes, assume there is just one node with whole RAM */
556 if (!nb_numa_nodes
) {
558 ramnode
.node_mem
= machine
->ram_size
;
562 for (i
= 0, mem_start
= 0; i
< nb_nodes
; ++i
) {
563 if (!nodes
[i
].node_mem
) {
566 if (mem_start
>= machine
->ram_size
) {
569 node_size
= nodes
[i
].node_mem
;
570 if (node_size
> machine
->ram_size
- mem_start
) {
571 node_size
= machine
->ram_size
- mem_start
;
575 /* ppc_spapr_init() checks for rma_size <= node0_size already */
576 spapr_populate_memory_node(fdt
, i
, 0, spapr
->rma_size
);
577 mem_start
+= spapr
->rma_size
;
578 node_size
-= spapr
->rma_size
;
580 for ( ; node_size
; ) {
581 hwaddr sizetmp
= pow2floor(node_size
);
583 /* mem_start != 0 here */
584 if (ctzl(mem_start
) < ctzl(sizetmp
)) {
585 sizetmp
= 1ULL << ctzl(mem_start
);
588 spapr_populate_memory_node(fdt
, i
, mem_start
, sizetmp
);
589 node_size
-= sizetmp
;
590 mem_start
+= sizetmp
;
597 static void spapr_populate_cpu_dt(CPUState
*cs
, void *fdt
, int offset
,
598 sPAPRMachineState
*spapr
)
600 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
601 CPUPPCState
*env
= &cpu
->env
;
602 PowerPCCPUClass
*pcc
= POWERPC_CPU_GET_CLASS(cs
);
603 int index
= ppc_get_vcpu_dt_id(cpu
);
604 uint32_t segs
[] = {cpu_to_be32(28), cpu_to_be32(40),
605 0xffffffff, 0xffffffff};
606 uint32_t tbfreq
= kvm_enabled() ? kvmppc_get_tbfreq()
607 : SPAPR_TIMEBASE_FREQ
;
608 uint32_t cpufreq
= kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
609 uint32_t page_sizes_prop
[64];
610 size_t page_sizes_prop_size
;
611 uint32_t vcpus_per_socket
= smp_threads
* smp_cores
;
612 uint32_t pft_size_prop
[] = {0, cpu_to_be32(spapr
->htab_shift
)};
613 sPAPRDRConnector
*drc
;
614 sPAPRDRConnectorClass
*drck
;
617 drc
= spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU
, index
);
619 drck
= SPAPR_DR_CONNECTOR_GET_CLASS(drc
);
620 drc_index
= drck
->get_index(drc
);
621 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,my-drc-index", drc_index
)));
624 /* Note: we keep CI large pages off for now because a 64K capable guest
625 * provisioned with large pages might otherwise try to map a qemu
626 * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
627 * even if that qemu runs on a 4k host.
629 * We can later add this bit back when we are confident this is not
630 * an issue (!HV KVM or 64K host)
632 uint8_t pa_features_206
[] = { 6, 0,
633 0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
634 uint8_t pa_features_207
[] = { 24, 0,
635 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
636 0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
637 0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
638 0x80, 0x00, 0x80, 0x00, 0x80, 0x00 };
639 uint8_t *pa_features
;
642 _FDT((fdt_setprop_cell(fdt
, offset
, "reg", index
)));
643 _FDT((fdt_setprop_string(fdt
, offset
, "device_type", "cpu")));
645 _FDT((fdt_setprop_cell(fdt
, offset
, "cpu-version", env
->spr
[SPR_PVR
])));
646 _FDT((fdt_setprop_cell(fdt
, offset
, "d-cache-block-size",
647 env
->dcache_line_size
)));
648 _FDT((fdt_setprop_cell(fdt
, offset
, "d-cache-line-size",
649 env
->dcache_line_size
)));
650 _FDT((fdt_setprop_cell(fdt
, offset
, "i-cache-block-size",
651 env
->icache_line_size
)));
652 _FDT((fdt_setprop_cell(fdt
, offset
, "i-cache-line-size",
653 env
->icache_line_size
)));
655 if (pcc
->l1_dcache_size
) {
656 _FDT((fdt_setprop_cell(fdt
, offset
, "d-cache-size",
657 pcc
->l1_dcache_size
)));
659 fprintf(stderr
, "Warning: Unknown L1 dcache size for cpu\n");
661 if (pcc
->l1_icache_size
) {
662 _FDT((fdt_setprop_cell(fdt
, offset
, "i-cache-size",
663 pcc
->l1_icache_size
)));
665 fprintf(stderr
, "Warning: Unknown L1 icache size for cpu\n");
668 _FDT((fdt_setprop_cell(fdt
, offset
, "timebase-frequency", tbfreq
)));
669 _FDT((fdt_setprop_cell(fdt
, offset
, "clock-frequency", cpufreq
)));
670 _FDT((fdt_setprop_cell(fdt
, offset
, "slb-size", env
->slb_nr
)));
671 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,slb-size", env
->slb_nr
)));
672 _FDT((fdt_setprop_string(fdt
, offset
, "status", "okay")));
673 _FDT((fdt_setprop(fdt
, offset
, "64-bit", NULL
, 0)));
675 if (env
->spr_cb
[SPR_PURR
].oea_read
) {
676 _FDT((fdt_setprop(fdt
, offset
, "ibm,purr", NULL
, 0)));
679 if (env
->mmu_model
& POWERPC_MMU_1TSEG
) {
680 _FDT((fdt_setprop(fdt
, offset
, "ibm,processor-segment-sizes",
681 segs
, sizeof(segs
))));
684 /* Advertise VMX/VSX (vector extensions) if available
685 * 0 / no property == no vector extensions
686 * 1 == VMX / Altivec available
687 * 2 == VSX available */
688 if (env
->insns_flags
& PPC_ALTIVEC
) {
689 uint32_t vmx
= (env
->insns_flags2
& PPC2_VSX
) ? 2 : 1;
691 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,vmx", vmx
)));
694 /* Advertise DFP (Decimal Floating Point) if available
695 * 0 / no property == no DFP
696 * 1 == DFP available */
697 if (env
->insns_flags2
& PPC2_DFP
) {
698 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,dfp", 1)));
701 page_sizes_prop_size
= create_page_sizes_prop(env
, page_sizes_prop
,
702 sizeof(page_sizes_prop
));
703 if (page_sizes_prop_size
) {
704 _FDT((fdt_setprop(fdt
, offset
, "ibm,segment-page-sizes",
705 page_sizes_prop
, page_sizes_prop_size
)));
708 /* Do the ibm,pa-features property, adjust it for ci-large-pages */
709 if (env
->mmu_model
== POWERPC_MMU_2_06
) {
710 pa_features
= pa_features_206
;
711 pa_size
= sizeof(pa_features_206
);
712 } else /* env->mmu_model == POWERPC_MMU_2_07 */ {
713 pa_features
= pa_features_207
;
714 pa_size
= sizeof(pa_features_207
);
716 if (env
->ci_large_pages
) {
717 pa_features
[3] |= 0x20;
719 _FDT((fdt_setprop(fdt
, offset
, "ibm,pa-features", pa_features
, pa_size
)));
721 _FDT((fdt_setprop_cell(fdt
, offset
, "ibm,chip-id",
722 cs
->cpu_index
/ vcpus_per_socket
)));
724 _FDT((fdt_setprop(fdt
, offset
, "ibm,pft-size",
725 pft_size_prop
, sizeof(pft_size_prop
))));
727 _FDT(spapr_fixup_cpu_numa_dt(fdt
, offset
, cs
));
729 _FDT(spapr_fixup_cpu_smt_dt(fdt
, offset
, cpu
,
730 ppc_get_compat_smt_threads(cpu
)));
733 static void spapr_populate_cpus_dt_node(void *fdt
, sPAPRMachineState
*spapr
)
738 int smt
= kvmppc_smt_threads();
740 cpus_offset
= fdt_add_subnode(fdt
, 0, "cpus");
742 _FDT((fdt_setprop_cell(fdt
, cpus_offset
, "#address-cells", 0x1)));
743 _FDT((fdt_setprop_cell(fdt
, cpus_offset
, "#size-cells", 0x0)));
746 * We walk the CPUs in reverse order to ensure that CPU DT nodes
747 * created by fdt_add_subnode() end up in the right order in FDT
748 * for the guest kernel the enumerate the CPUs correctly.
750 CPU_FOREACH_REVERSE(cs
) {
751 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
752 int index
= ppc_get_vcpu_dt_id(cpu
);
753 DeviceClass
*dc
= DEVICE_GET_CLASS(cs
);
756 if ((index
% smt
) != 0) {
760 nodename
= g_strdup_printf("%s@%x", dc
->fw_name
, index
);
761 offset
= fdt_add_subnode(fdt
, cpus_offset
, nodename
);
764 spapr_populate_cpu_dt(cs
, fdt
, offset
, spapr
);
770 * Adds ibm,dynamic-reconfiguration-memory node.
771 * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
772 * of this device tree node.
774 static int spapr_populate_drconf_memory(sPAPRMachineState
*spapr
, void *fdt
)
776 MachineState
*machine
= MACHINE(spapr
);
778 uint64_t lmb_size
= SPAPR_MEMORY_BLOCK_SIZE
;
779 uint32_t prop_lmb_size
[] = {0, cpu_to_be32(lmb_size
)};
780 uint32_t hotplug_lmb_start
= spapr
->hotplug_memory
.base
/ lmb_size
;
781 uint32_t nr_lmbs
= (spapr
->hotplug_memory
.base
+
782 memory_region_size(&spapr
->hotplug_memory
.mr
)) /
784 uint32_t *int_buf
, *cur_index
, buf_len
;
785 int nr_nodes
= nb_numa_nodes
? nb_numa_nodes
: 1;
788 * Don't create the node if there is no hotpluggable memory
790 if (machine
->ram_size
== machine
->maxram_size
) {
795 * Allocate enough buffer size to fit in ibm,dynamic-memory
796 * or ibm,associativity-lookup-arrays
798 buf_len
= MAX(nr_lmbs
* SPAPR_DR_LMB_LIST_ENTRY_SIZE
+ 1, nr_nodes
* 4 + 2)
800 cur_index
= int_buf
= g_malloc0(buf_len
);
802 offset
= fdt_add_subnode(fdt
, 0, "ibm,dynamic-reconfiguration-memory");
804 ret
= fdt_setprop(fdt
, offset
, "ibm,lmb-size", prop_lmb_size
,
805 sizeof(prop_lmb_size
));
810 ret
= fdt_setprop_cell(fdt
, offset
, "ibm,memory-flags-mask", 0xff);
815 ret
= fdt_setprop_cell(fdt
, offset
, "ibm,memory-preservation-time", 0x0);
820 /* ibm,dynamic-memory */
821 int_buf
[0] = cpu_to_be32(nr_lmbs
);
823 for (i
= 0; i
< nr_lmbs
; i
++) {
824 uint64_t addr
= i
* lmb_size
;
825 uint32_t *dynamic_memory
= cur_index
;
827 if (i
>= hotplug_lmb_start
) {
828 sPAPRDRConnector
*drc
;
829 sPAPRDRConnectorClass
*drck
;
831 drc
= spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB
, i
);
833 drck
= SPAPR_DR_CONNECTOR_GET_CLASS(drc
);
835 dynamic_memory
[0] = cpu_to_be32(addr
>> 32);
836 dynamic_memory
[1] = cpu_to_be32(addr
& 0xffffffff);
837 dynamic_memory
[2] = cpu_to_be32(drck
->get_index(drc
));
838 dynamic_memory
[3] = cpu_to_be32(0); /* reserved */
839 dynamic_memory
[4] = cpu_to_be32(numa_get_node(addr
, NULL
));
840 if (memory_region_present(get_system_memory(), addr
)) {
841 dynamic_memory
[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED
);
843 dynamic_memory
[5] = cpu_to_be32(0);
847 * LMB information for RMA, boot time RAM and gap b/n RAM and
848 * hotplug memory region -- all these are marked as reserved
849 * and as having no valid DRC.
851 dynamic_memory
[0] = cpu_to_be32(addr
>> 32);
852 dynamic_memory
[1] = cpu_to_be32(addr
& 0xffffffff);
853 dynamic_memory
[2] = cpu_to_be32(0);
854 dynamic_memory
[3] = cpu_to_be32(0); /* reserved */
855 dynamic_memory
[4] = cpu_to_be32(-1);
856 dynamic_memory
[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED
|
857 SPAPR_LMB_FLAGS_DRC_INVALID
);
860 cur_index
+= SPAPR_DR_LMB_LIST_ENTRY_SIZE
;
862 ret
= fdt_setprop(fdt
, offset
, "ibm,dynamic-memory", int_buf
, buf_len
);
867 /* ibm,associativity-lookup-arrays */
869 int_buf
[0] = cpu_to_be32(nr_nodes
);
870 int_buf
[1] = cpu_to_be32(4); /* Number of entries per associativity list */
872 for (i
= 0; i
< nr_nodes
; i
++) {
873 uint32_t associativity
[] = {
879 memcpy(cur_index
, associativity
, sizeof(associativity
));
882 ret
= fdt_setprop(fdt
, offset
, "ibm,associativity-lookup-arrays", int_buf
,
883 (cur_index
- int_buf
) * sizeof(uint32_t));
889 int spapr_h_cas_compose_response(sPAPRMachineState
*spapr
,
890 target_ulong addr
, target_ulong size
,
891 bool cpu_update
, bool memory_update
)
893 void *fdt
, *fdt_skel
;
894 sPAPRDeviceTreeUpdateHeader hdr
= { .version_id
= 1 };
895 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
899 /* Create sceleton */
900 fdt_skel
= g_malloc0(size
);
901 _FDT((fdt_create(fdt_skel
, size
)));
902 _FDT((fdt_begin_node(fdt_skel
, "")));
903 _FDT((fdt_end_node(fdt_skel
)));
904 _FDT((fdt_finish(fdt_skel
)));
905 fdt
= g_malloc0(size
);
906 _FDT((fdt_open_into(fdt_skel
, fdt
, size
)));
909 /* Fixup cpu nodes */
911 _FDT((spapr_fixup_cpu_dt(fdt
, spapr
)));
914 /* Generate ibm,dynamic-reconfiguration-memory node if required */
915 if (memory_update
&& smc
->dr_lmb_enabled
) {
916 _FDT((spapr_populate_drconf_memory(spapr
, fdt
)));
919 /* Pack resulting tree */
920 _FDT((fdt_pack(fdt
)));
922 if (fdt_totalsize(fdt
) + sizeof(hdr
) > size
) {
923 trace_spapr_cas_failed(size
);
927 cpu_physical_memory_write(addr
, &hdr
, sizeof(hdr
));
928 cpu_physical_memory_write(addr
+ sizeof(hdr
), fdt
, fdt_totalsize(fdt
));
929 trace_spapr_cas_continue(fdt_totalsize(fdt
) + sizeof(hdr
));
935 static void spapr_finalize_fdt(sPAPRMachineState
*spapr
,
940 MachineState
*machine
= MACHINE(qdev_get_machine());
941 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(machine
);
942 const char *boot_device
= machine
->boot_order
;
949 fdt
= g_malloc(FDT_MAX_SIZE
);
951 /* open out the base tree into a temp buffer for the final tweaks */
952 _FDT((fdt_open_into(spapr
->fdt_skel
, fdt
, FDT_MAX_SIZE
)));
954 ret
= spapr_populate_memory(spapr
, fdt
);
956 fprintf(stderr
, "couldn't setup memory nodes in fdt\n");
960 ret
= spapr_populate_vdevice(spapr
->vio_bus
, fdt
);
962 fprintf(stderr
, "couldn't setup vio devices in fdt\n");
966 if (object_resolve_path_type("", TYPE_SPAPR_RNG
, NULL
)) {
967 ret
= spapr_rng_populate_dt(fdt
);
969 fprintf(stderr
, "could not set up rng device in the fdt\n");
974 QLIST_FOREACH(phb
, &spapr
->phbs
, list
) {
975 ret
= spapr_populate_pci_dt(phb
, PHANDLE_XICP
, fdt
);
977 error_report("couldn't setup PCI devices in fdt");
983 ret
= spapr_rtas_device_tree_setup(fdt
, rtas_addr
, rtas_size
);
985 fprintf(stderr
, "Couldn't set up RTAS device tree properties\n");
989 spapr_populate_cpus_dt_node(fdt
, spapr
);
991 bootlist
= get_boot_devices_list(&cb
, true);
992 if (cb
&& bootlist
) {
993 int offset
= fdt_path_offset(fdt
, "/chosen");
997 for (i
= 0; i
< cb
; i
++) {
998 if (bootlist
[i
] == '\n') {
1003 ret
= fdt_setprop_string(fdt
, offset
, "qemu,boot-list", bootlist
);
1006 if (boot_device
&& strlen(boot_device
)) {
1007 int offset
= fdt_path_offset(fdt
, "/chosen");
1012 fdt_setprop_string(fdt
, offset
, "qemu,boot-device", boot_device
);
1015 if (!spapr
->has_graphics
) {
1016 spapr_populate_chosen_stdout(fdt
, spapr
->vio_bus
);
1019 if (smc
->dr_lmb_enabled
) {
1020 _FDT(spapr_drc_populate_dt(fdt
, 0, NULL
, SPAPR_DR_CONNECTOR_TYPE_LMB
));
1023 if (smc
->dr_cpu_enabled
) {
1024 int offset
= fdt_path_offset(fdt
, "/cpus");
1025 ret
= spapr_drc_populate_dt(fdt
, offset
, NULL
,
1026 SPAPR_DR_CONNECTOR_TYPE_CPU
);
1028 error_report("Couldn't set up CPU DR device tree properties");
1033 _FDT((fdt_pack(fdt
)));
1035 if (fdt_totalsize(fdt
) > FDT_MAX_SIZE
) {
1036 error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1037 fdt_totalsize(fdt
), FDT_MAX_SIZE
);
1041 qemu_fdt_dumpdtb(fdt
, fdt_totalsize(fdt
));
1042 cpu_physical_memory_write(fdt_addr
, fdt
, fdt_totalsize(fdt
));
1048 static uint64_t translate_kernel_address(void *opaque
, uint64_t addr
)
1050 return (addr
& 0x0fffffff) + KERNEL_LOAD_ADDR
;
1053 static void emulate_spapr_hypercall(PowerPCCPU
*cpu
)
1055 CPUPPCState
*env
= &cpu
->env
;
1058 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1059 env
->gpr
[3] = H_PRIVILEGE
;
1061 env
->gpr
[3] = spapr_hypercall(cpu
, env
->gpr
[3], &env
->gpr
[4]);
1065 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1066 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1067 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1068 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1069 #define DIRTY_HPTE(_hpte) ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1072 * Get the fd to access the kernel htab, re-opening it if necessary
1074 static int get_htab_fd(sPAPRMachineState
*spapr
)
1076 if (spapr
->htab_fd
>= 0) {
1077 return spapr
->htab_fd
;
1080 spapr
->htab_fd
= kvmppc_get_htab_fd(false);
1081 if (spapr
->htab_fd
< 0) {
1082 error_report("Unable to open fd for reading hash table from KVM: %s",
1086 return spapr
->htab_fd
;
1089 static void close_htab_fd(sPAPRMachineState
*spapr
)
1091 if (spapr
->htab_fd
>= 0) {
1092 close(spapr
->htab_fd
);
1094 spapr
->htab_fd
= -1;
1097 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize
)
1101 /* We aim for a hash table of size 1/128 the size of RAM (rounded
1102 * up). The PAPR recommendation is actually 1/64 of RAM size, but
1103 * that's much more than is needed for Linux guests */
1104 shift
= ctz64(pow2ceil(ramsize
)) - 7;
1105 shift
= MAX(shift
, 18); /* Minimum architected size */
1106 shift
= MIN(shift
, 46); /* Maximum architected size */
1110 static void spapr_reallocate_hpt(sPAPRMachineState
*spapr
, int shift
,
1115 /* Clean up any HPT info from a previous boot */
1116 g_free(spapr
->htab
);
1118 spapr
->htab_shift
= 0;
1119 close_htab_fd(spapr
);
1121 rc
= kvmppc_reset_htab(shift
);
1123 /* kernel-side HPT needed, but couldn't allocate one */
1124 error_setg_errno(errp
, errno
,
1125 "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1127 /* This is almost certainly fatal, but if the caller really
1128 * wants to carry on with shift == 0, it's welcome to try */
1129 } else if (rc
> 0) {
1130 /* kernel-side HPT allocated */
1133 "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1137 spapr
->htab_shift
= shift
;
1140 /* kernel-side HPT not needed, allocate in userspace instead */
1141 size_t size
= 1ULL << shift
;
1144 spapr
->htab
= qemu_memalign(size
, size
);
1146 error_setg_errno(errp
, errno
,
1147 "Could not allocate HPT of order %d", shift
);
1151 memset(spapr
->htab
, 0, size
);
1152 spapr
->htab_shift
= shift
;
1154 for (i
= 0; i
< size
/ HASH_PTE_SIZE_64
; i
++) {
1155 DIRTY_HPTE(HPTE(spapr
->htab
, i
));
1160 static int find_unknown_sysbus_device(SysBusDevice
*sbdev
, void *opaque
)
1162 bool matched
= false;
1164 if (object_dynamic_cast(OBJECT(sbdev
), TYPE_SPAPR_PCI_HOST_BRIDGE
)) {
1169 error_report("Device %s is not supported by this machine yet.",
1170 qdev_fw_name(DEVICE(sbdev
)));
1177 static void ppc_spapr_reset(void)
1179 MachineState
*machine
= MACHINE(qdev_get_machine());
1180 sPAPRMachineState
*spapr
= SPAPR_MACHINE(machine
);
1181 PowerPCCPU
*first_ppc_cpu
;
1182 uint32_t rtas_limit
;
1184 /* Check for unknown sysbus devices */
1185 foreach_dynamic_sysbus_device(find_unknown_sysbus_device
, NULL
);
1187 /* Allocate and/or reset the hash page table */
1188 spapr_reallocate_hpt(spapr
,
1189 spapr_hpt_shift_for_ramsize(machine
->maxram_size
),
1192 /* Update the RMA size if necessary */
1193 if (spapr
->vrma_adjust
) {
1194 spapr
->rma_size
= kvmppc_rma_size(spapr_node0_size(),
1198 qemu_devices_reset();
1201 * We place the device tree and RTAS just below either the top of the RMA,
1202 * or just below 2GB, whichever is lowere, so that it can be
1203 * processed with 32-bit real mode code if necessary
1205 rtas_limit
= MIN(spapr
->rma_size
, RTAS_MAX_ADDR
);
1206 spapr
->rtas_addr
= rtas_limit
- RTAS_MAX_SIZE
;
1207 spapr
->fdt_addr
= spapr
->rtas_addr
- FDT_MAX_SIZE
;
1210 spapr_finalize_fdt(spapr
, spapr
->fdt_addr
, spapr
->rtas_addr
,
1213 /* Copy RTAS over */
1214 cpu_physical_memory_write(spapr
->rtas_addr
, spapr
->rtas_blob
,
1217 /* Set up the entry state */
1218 first_ppc_cpu
= POWERPC_CPU(first_cpu
);
1219 first_ppc_cpu
->env
.gpr
[3] = spapr
->fdt_addr
;
1220 first_ppc_cpu
->env
.gpr
[5] = 0;
1221 first_cpu
->halted
= 0;
1222 first_ppc_cpu
->env
.nip
= SPAPR_ENTRY_POINT
;
1226 static void spapr_create_nvram(sPAPRMachineState
*spapr
)
1228 DeviceState
*dev
= qdev_create(&spapr
->vio_bus
->bus
, "spapr-nvram");
1229 DriveInfo
*dinfo
= drive_get(IF_PFLASH
, 0, 0);
1232 qdev_prop_set_drive(dev
, "drive", blk_by_legacy_dinfo(dinfo
),
1236 qdev_init_nofail(dev
);
1238 spapr
->nvram
= (struct sPAPRNVRAM
*)dev
;
1241 static void spapr_rtc_create(sPAPRMachineState
*spapr
)
1243 DeviceState
*dev
= qdev_create(NULL
, TYPE_SPAPR_RTC
);
1245 qdev_init_nofail(dev
);
1248 object_property_add_alias(qdev_get_machine(), "rtc-time",
1249 OBJECT(spapr
->rtc
), "date", NULL
);
1252 /* Returns whether we want to use VGA or not */
1253 static bool spapr_vga_init(PCIBus
*pci_bus
, Error
**errp
)
1255 switch (vga_interface_type
) {
1262 return pci_vga_init(pci_bus
) != NULL
;
1265 "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1270 static int spapr_post_load(void *opaque
, int version_id
)
1272 sPAPRMachineState
*spapr
= (sPAPRMachineState
*)opaque
;
1275 /* In earlier versions, there was no separate qdev for the PAPR
1276 * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1277 * So when migrating from those versions, poke the incoming offset
1278 * value into the RTC device */
1279 if (version_id
< 3) {
1280 err
= spapr_rtc_import_offset(spapr
->rtc
, spapr
->rtc_offset
);
1286 static bool version_before_3(void *opaque
, int version_id
)
1288 return version_id
< 3;
1291 static const VMStateDescription vmstate_spapr
= {
1294 .minimum_version_id
= 1,
1295 .post_load
= spapr_post_load
,
1296 .fields
= (VMStateField
[]) {
1297 /* used to be @next_irq */
1298 VMSTATE_UNUSED_BUFFER(version_before_3
, 0, 4),
1301 VMSTATE_UINT64_TEST(rtc_offset
, sPAPRMachineState
, version_before_3
),
1303 VMSTATE_PPC_TIMEBASE_V(tb
, sPAPRMachineState
, 2),
1304 VMSTATE_END_OF_LIST()
1308 static int htab_save_setup(QEMUFile
*f
, void *opaque
)
1310 sPAPRMachineState
*spapr
= opaque
;
1312 /* "Iteration" header */
1313 qemu_put_be32(f
, spapr
->htab_shift
);
1316 spapr
->htab_save_index
= 0;
1317 spapr
->htab_first_pass
= true;
1319 assert(kvm_enabled());
1326 static void htab_save_first_pass(QEMUFile
*f
, sPAPRMachineState
*spapr
,
1329 bool has_timeout
= max_ns
!= -1;
1330 int htabslots
= HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
;
1331 int index
= spapr
->htab_save_index
;
1332 int64_t starttime
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
1334 assert(spapr
->htab_first_pass
);
1339 /* Consume invalid HPTEs */
1340 while ((index
< htabslots
)
1341 && !HPTE_VALID(HPTE(spapr
->htab
, index
))) {
1343 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
1346 /* Consume valid HPTEs */
1348 while ((index
< htabslots
) && (index
- chunkstart
< USHRT_MAX
)
1349 && HPTE_VALID(HPTE(spapr
->htab
, index
))) {
1351 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
1354 if (index
> chunkstart
) {
1355 int n_valid
= index
- chunkstart
;
1357 qemu_put_be32(f
, chunkstart
);
1358 qemu_put_be16(f
, n_valid
);
1359 qemu_put_be16(f
, 0);
1360 qemu_put_buffer(f
, HPTE(spapr
->htab
, chunkstart
),
1361 HASH_PTE_SIZE_64
* n_valid
);
1364 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - starttime
) > max_ns
) {
1368 } while ((index
< htabslots
) && !qemu_file_rate_limit(f
));
1370 if (index
>= htabslots
) {
1371 assert(index
== htabslots
);
1373 spapr
->htab_first_pass
= false;
1375 spapr
->htab_save_index
= index
;
1378 static int htab_save_later_pass(QEMUFile
*f
, sPAPRMachineState
*spapr
,
1381 bool final
= max_ns
< 0;
1382 int htabslots
= HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
;
1383 int examined
= 0, sent
= 0;
1384 int index
= spapr
->htab_save_index
;
1385 int64_t starttime
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
1387 assert(!spapr
->htab_first_pass
);
1390 int chunkstart
, invalidstart
;
1392 /* Consume non-dirty HPTEs */
1393 while ((index
< htabslots
)
1394 && !HPTE_DIRTY(HPTE(spapr
->htab
, index
))) {
1400 /* Consume valid dirty HPTEs */
1401 while ((index
< htabslots
) && (index
- chunkstart
< USHRT_MAX
)
1402 && HPTE_DIRTY(HPTE(spapr
->htab
, index
))
1403 && HPTE_VALID(HPTE(spapr
->htab
, index
))) {
1404 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
1409 invalidstart
= index
;
1410 /* Consume invalid dirty HPTEs */
1411 while ((index
< htabslots
) && (index
- invalidstart
< USHRT_MAX
)
1412 && HPTE_DIRTY(HPTE(spapr
->htab
, index
))
1413 && !HPTE_VALID(HPTE(spapr
->htab
, index
))) {
1414 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
1419 if (index
> chunkstart
) {
1420 int n_valid
= invalidstart
- chunkstart
;
1421 int n_invalid
= index
- invalidstart
;
1423 qemu_put_be32(f
, chunkstart
);
1424 qemu_put_be16(f
, n_valid
);
1425 qemu_put_be16(f
, n_invalid
);
1426 qemu_put_buffer(f
, HPTE(spapr
->htab
, chunkstart
),
1427 HASH_PTE_SIZE_64
* n_valid
);
1428 sent
+= index
- chunkstart
;
1430 if (!final
&& (qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - starttime
) > max_ns
) {
1435 if (examined
>= htabslots
) {
1439 if (index
>= htabslots
) {
1440 assert(index
== htabslots
);
1443 } while ((examined
< htabslots
) && (!qemu_file_rate_limit(f
) || final
));
1445 if (index
>= htabslots
) {
1446 assert(index
== htabslots
);
1450 spapr
->htab_save_index
= index
;
1452 return (examined
>= htabslots
) && (sent
== 0) ? 1 : 0;
1455 #define MAX_ITERATION_NS 5000000 /* 5 ms */
1456 #define MAX_KVM_BUF_SIZE 2048
1458 static int htab_save_iterate(QEMUFile
*f
, void *opaque
)
1460 sPAPRMachineState
*spapr
= opaque
;
1464 /* Iteration header */
1465 qemu_put_be32(f
, 0);
1468 assert(kvm_enabled());
1470 fd
= get_htab_fd(spapr
);
1475 rc
= kvmppc_save_htab(f
, fd
, MAX_KVM_BUF_SIZE
, MAX_ITERATION_NS
);
1479 } else if (spapr
->htab_first_pass
) {
1480 htab_save_first_pass(f
, spapr
, MAX_ITERATION_NS
);
1482 rc
= htab_save_later_pass(f
, spapr
, MAX_ITERATION_NS
);
1486 qemu_put_be32(f
, 0);
1487 qemu_put_be16(f
, 0);
1488 qemu_put_be16(f
, 0);
1493 static int htab_save_complete(QEMUFile
*f
, void *opaque
)
1495 sPAPRMachineState
*spapr
= opaque
;
1498 /* Iteration header */
1499 qemu_put_be32(f
, 0);
1504 assert(kvm_enabled());
1506 fd
= get_htab_fd(spapr
);
1511 rc
= kvmppc_save_htab(f
, fd
, MAX_KVM_BUF_SIZE
, -1);
1515 close_htab_fd(spapr
);
1517 if (spapr
->htab_first_pass
) {
1518 htab_save_first_pass(f
, spapr
, -1);
1520 htab_save_later_pass(f
, spapr
, -1);
1524 qemu_put_be32(f
, 0);
1525 qemu_put_be16(f
, 0);
1526 qemu_put_be16(f
, 0);
1531 static int htab_load(QEMUFile
*f
, void *opaque
, int version_id
)
1533 sPAPRMachineState
*spapr
= opaque
;
1534 uint32_t section_hdr
;
1537 if (version_id
< 1 || version_id
> 1) {
1538 error_report("htab_load() bad version");
1542 section_hdr
= qemu_get_be32(f
);
1545 Error
*local_err
= NULL
;
1547 /* First section gives the htab size */
1548 spapr_reallocate_hpt(spapr
, section_hdr
, &local_err
);
1550 error_report_err(local_err
);
1557 assert(kvm_enabled());
1559 fd
= kvmppc_get_htab_fd(true);
1561 error_report("Unable to open fd to restore KVM hash table: %s",
1568 uint16_t n_valid
, n_invalid
;
1570 index
= qemu_get_be32(f
);
1571 n_valid
= qemu_get_be16(f
);
1572 n_invalid
= qemu_get_be16(f
);
1574 if ((index
== 0) && (n_valid
== 0) && (n_invalid
== 0)) {
1579 if ((index
+ n_valid
+ n_invalid
) >
1580 (HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
)) {
1581 /* Bad index in stream */
1583 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1584 index
, n_valid
, n_invalid
, spapr
->htab_shift
);
1590 qemu_get_buffer(f
, HPTE(spapr
->htab
, index
),
1591 HASH_PTE_SIZE_64
* n_valid
);
1594 memset(HPTE(spapr
->htab
, index
+ n_valid
), 0,
1595 HASH_PTE_SIZE_64
* n_invalid
);
1602 rc
= kvmppc_load_htab_chunk(f
, fd
, index
, n_valid
, n_invalid
);
1617 static SaveVMHandlers savevm_htab_handlers
= {
1618 .save_live_setup
= htab_save_setup
,
1619 .save_live_iterate
= htab_save_iterate
,
1620 .save_live_complete_precopy
= htab_save_complete
,
1621 .load_state
= htab_load
,
1624 static void spapr_boot_set(void *opaque
, const char *boot_device
,
1627 MachineState
*machine
= MACHINE(qdev_get_machine());
1628 machine
->boot_order
= g_strdup(boot_device
);
1632 * Reset routine for LMB DR devices.
1634 * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1635 * routine. Reset for PCI DR devices will be handled by PHB reset routine
1636 * when it walks all its children devices. LMB devices reset occurs
1637 * as part of spapr_ppc_reset().
1639 static void spapr_drc_reset(void *opaque
)
1641 sPAPRDRConnector
*drc
= opaque
;
1642 DeviceState
*d
= DEVICE(drc
);
1649 static void spapr_create_lmb_dr_connectors(sPAPRMachineState
*spapr
)
1651 MachineState
*machine
= MACHINE(spapr
);
1652 uint64_t lmb_size
= SPAPR_MEMORY_BLOCK_SIZE
;
1653 uint32_t nr_lmbs
= (machine
->maxram_size
- machine
->ram_size
)/lmb_size
;
1656 for (i
= 0; i
< nr_lmbs
; i
++) {
1657 sPAPRDRConnector
*drc
;
1660 addr
= i
* lmb_size
+ spapr
->hotplug_memory
.base
;
1661 drc
= spapr_dr_connector_new(OBJECT(spapr
), SPAPR_DR_CONNECTOR_TYPE_LMB
,
1663 qemu_register_reset(spapr_drc_reset
, drc
);
1668 * If RAM size, maxmem size and individual node mem sizes aren't aligned
1669 * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1670 * since we can't support such unaligned sizes with DRCONF_MEMORY.
1672 static void spapr_validate_node_memory(MachineState
*machine
, Error
**errp
)
1676 if (machine
->ram_size
% SPAPR_MEMORY_BLOCK_SIZE
) {
1677 error_setg(errp
, "Memory size 0x" RAM_ADDR_FMT
1678 " is not aligned to %llu MiB",
1680 SPAPR_MEMORY_BLOCK_SIZE
/ M_BYTE
);
1684 if (machine
->maxram_size
% SPAPR_MEMORY_BLOCK_SIZE
) {
1685 error_setg(errp
, "Maximum memory size 0x" RAM_ADDR_FMT
1686 " is not aligned to %llu MiB",
1688 SPAPR_MEMORY_BLOCK_SIZE
/ M_BYTE
);
1692 for (i
= 0; i
< nb_numa_nodes
; i
++) {
1693 if (numa_info
[i
].node_mem
% SPAPR_MEMORY_BLOCK_SIZE
) {
1695 "Node %d memory size 0x%" PRIx64
1696 " is not aligned to %llu MiB",
1697 i
, numa_info
[i
].node_mem
,
1698 SPAPR_MEMORY_BLOCK_SIZE
/ M_BYTE
);
1704 /* pSeries LPAR / sPAPR hardware init */
1705 static void ppc_spapr_init(MachineState
*machine
)
1707 sPAPRMachineState
*spapr
= SPAPR_MACHINE(machine
);
1708 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(machine
);
1709 const char *kernel_filename
= machine
->kernel_filename
;
1710 const char *kernel_cmdline
= machine
->kernel_cmdline
;
1711 const char *initrd_filename
= machine
->initrd_filename
;
1714 MemoryRegion
*sysmem
= get_system_memory();
1715 MemoryRegion
*ram
= g_new(MemoryRegion
, 1);
1716 MemoryRegion
*rma_region
;
1718 hwaddr rma_alloc_size
;
1719 hwaddr node0_size
= spapr_node0_size();
1720 uint32_t initrd_base
= 0;
1721 long kernel_size
= 0, initrd_size
= 0;
1722 long load_limit
, fw_size
;
1723 bool kernel_le
= false;
1725 int smt
= kvmppc_smt_threads();
1726 int spapr_cores
= smp_cpus
/ smp_threads
;
1727 int spapr_max_cores
= max_cpus
/ smp_threads
;
1729 if (smc
->dr_cpu_enabled
) {
1730 if (smp_cpus
% smp_threads
) {
1731 error_report("smp_cpus (%u) must be multiple of threads (%u)",
1732 smp_cpus
, smp_threads
);
1735 if (max_cpus
% smp_threads
) {
1736 error_report("max_cpus (%u) must be multiple of threads (%u)",
1737 max_cpus
, smp_threads
);
1742 msi_nonbroken
= true;
1744 QLIST_INIT(&spapr
->phbs
);
1746 cpu_ppc_hypercall
= emulate_spapr_hypercall
;
1748 /* Allocate RMA if necessary */
1749 rma_alloc_size
= kvmppc_alloc_rma(&rma
);
1751 if (rma_alloc_size
== -1) {
1752 error_report("Unable to create RMA");
1756 if (rma_alloc_size
&& (rma_alloc_size
< node0_size
)) {
1757 spapr
->rma_size
= rma_alloc_size
;
1759 spapr
->rma_size
= node0_size
;
1761 /* With KVM, we don't actually know whether KVM supports an
1762 * unbounded RMA (PR KVM) or is limited by the hash table size
1763 * (HV KVM using VRMA), so we always assume the latter
1765 * In that case, we also limit the initial allocations for RTAS
1766 * etc... to 256M since we have no way to know what the VRMA size
1767 * is going to be as it depends on the size of the hash table
1768 * isn't determined yet.
1770 if (kvm_enabled()) {
1771 spapr
->vrma_adjust
= 1;
1772 spapr
->rma_size
= MIN(spapr
->rma_size
, 0x10000000);
1775 /* Actually we don't support unbounded RMA anymore since we
1776 * added proper emulation of HV mode. The max we can get is
1777 * 16G which also happens to be what we configure for PAPR
1778 * mode so make sure we don't do anything bigger than that
1780 spapr
->rma_size
= MIN(spapr
->rma_size
, 0x400000000ull
);
1783 if (spapr
->rma_size
> node0_size
) {
1784 error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx
")",
1789 /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1790 load_limit
= MIN(spapr
->rma_size
, RTAS_MAX_ADDR
) - FW_OVERHEAD
;
1792 /* Set up Interrupt Controller before we create the VCPUs */
1793 spapr
->xics
= xics_system_init(machine
,
1794 DIV_ROUND_UP(max_cpus
* smt
, smp_threads
),
1795 XICS_IRQS_SPAPR
, &error_fatal
);
1797 if (smc
->dr_lmb_enabled
) {
1798 spapr_validate_node_memory(machine
, &error_fatal
);
1802 if (machine
->cpu_model
== NULL
) {
1803 machine
->cpu_model
= kvm_enabled() ? "host" : "POWER7";
1806 if (smc
->dr_cpu_enabled
) {
1807 char *type
= spapr_get_cpu_core_type(machine
->cpu_model
);
1809 spapr
->cores
= g_new0(Object
*, spapr_max_cores
);
1810 for (i
= 0; i
< spapr_max_cores
; i
++) {
1811 int core_dt_id
= i
* smt
;
1812 sPAPRDRConnector
*drc
=
1813 spapr_dr_connector_new(OBJECT(spapr
),
1814 SPAPR_DR_CONNECTOR_TYPE_CPU
, core_dt_id
);
1816 qemu_register_reset(spapr_drc_reset
, drc
);
1818 if (i
< spapr_cores
) {
1819 char *type
= spapr_get_cpu_core_type(machine
->cpu_model
);
1822 if (!object_class_by_name(type
)) {
1823 error_report("Unable to find sPAPR CPU Core definition");
1827 core
= object_new(type
);
1828 object_property_set_int(core
, smp_threads
, "nr-threads",
1830 object_property_set_int(core
, core_dt_id
, CPU_CORE_PROP_CORE_ID
,
1832 object_property_set_bool(core
, true, "realized", &error_fatal
);
1837 for (i
= 0; i
< smp_cpus
; i
++) {
1838 PowerPCCPU
*cpu
= cpu_ppc_init(machine
->cpu_model
);
1840 error_report("Unable to find PowerPC CPU definition");
1843 spapr_cpu_init(spapr
, cpu
, &error_fatal
);
1847 if (kvm_enabled()) {
1848 /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
1849 kvmppc_enable_logical_ci_hcalls();
1850 kvmppc_enable_set_mode_hcall();
1854 memory_region_allocate_system_memory(ram
, NULL
, "ppc_spapr.ram",
1856 memory_region_add_subregion(sysmem
, 0, ram
);
1858 if (rma_alloc_size
&& rma
) {
1859 rma_region
= g_new(MemoryRegion
, 1);
1860 memory_region_init_ram_ptr(rma_region
, NULL
, "ppc_spapr.rma",
1861 rma_alloc_size
, rma
);
1862 vmstate_register_ram_global(rma_region
);
1863 memory_region_add_subregion(sysmem
, 0, rma_region
);
1866 /* initialize hotplug memory address space */
1867 if (machine
->ram_size
< machine
->maxram_size
) {
1868 ram_addr_t hotplug_mem_size
= machine
->maxram_size
- machine
->ram_size
;
1870 * Limit the number of hotpluggable memory slots to half the number
1871 * slots that KVM supports, leaving the other half for PCI and other
1872 * devices. However ensure that number of slots doesn't drop below 32.
1874 int max_memslots
= kvm_enabled() ? kvm_get_max_memslots() / 2 :
1875 SPAPR_MAX_RAM_SLOTS
;
1877 if (max_memslots
< SPAPR_MAX_RAM_SLOTS
) {
1878 max_memslots
= SPAPR_MAX_RAM_SLOTS
;
1880 if (machine
->ram_slots
> max_memslots
) {
1881 error_report("Specified number of memory slots %"
1882 PRIu64
" exceeds max supported %d",
1883 machine
->ram_slots
, max_memslots
);
1887 spapr
->hotplug_memory
.base
= ROUND_UP(machine
->ram_size
,
1888 SPAPR_HOTPLUG_MEM_ALIGN
);
1889 memory_region_init(&spapr
->hotplug_memory
.mr
, OBJECT(spapr
),
1890 "hotplug-memory", hotplug_mem_size
);
1891 memory_region_add_subregion(sysmem
, spapr
->hotplug_memory
.base
,
1892 &spapr
->hotplug_memory
.mr
);
1895 if (smc
->dr_lmb_enabled
) {
1896 spapr_create_lmb_dr_connectors(spapr
);
1899 filename
= qemu_find_file(QEMU_FILE_TYPE_BIOS
, "spapr-rtas.bin");
1901 error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
1904 spapr
->rtas_size
= get_image_size(filename
);
1905 if (spapr
->rtas_size
< 0) {
1906 error_report("Could not get size of LPAR rtas '%s'", filename
);
1909 spapr
->rtas_blob
= g_malloc(spapr
->rtas_size
);
1910 if (load_image_size(filename
, spapr
->rtas_blob
, spapr
->rtas_size
) < 0) {
1911 error_report("Could not load LPAR rtas '%s'", filename
);
1914 if (spapr
->rtas_size
> RTAS_MAX_SIZE
) {
1915 error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
1916 (size_t)spapr
->rtas_size
, RTAS_MAX_SIZE
);
1921 /* Set up EPOW events infrastructure */
1922 spapr_events_init(spapr
);
1924 /* Set up the RTC RTAS interfaces */
1925 spapr_rtc_create(spapr
);
1927 /* Set up VIO bus */
1928 spapr
->vio_bus
= spapr_vio_bus_init();
1930 for (i
= 0; i
< MAX_SERIAL_PORTS
; i
++) {
1931 if (serial_hds
[i
]) {
1932 spapr_vty_create(spapr
->vio_bus
, serial_hds
[i
]);
1936 /* We always have at least the nvram device on VIO */
1937 spapr_create_nvram(spapr
);
1940 spapr_pci_rtas_init();
1942 phb
= spapr_create_phb(spapr
, 0);
1944 for (i
= 0; i
< nb_nics
; i
++) {
1945 NICInfo
*nd
= &nd_table
[i
];
1948 nd
->model
= g_strdup("ibmveth");
1951 if (strcmp(nd
->model
, "ibmveth") == 0) {
1952 spapr_vlan_create(spapr
->vio_bus
, nd
);
1954 pci_nic_init_nofail(&nd_table
[i
], phb
->bus
, nd
->model
, NULL
);
1958 for (i
= 0; i
<= drive_get_max_bus(IF_SCSI
); i
++) {
1959 spapr_vscsi_create(spapr
->vio_bus
);
1963 if (spapr_vga_init(phb
->bus
, &error_fatal
)) {
1964 spapr
->has_graphics
= true;
1965 machine
->usb
|= defaults_enabled() && !machine
->usb_disabled
;
1969 if (smc
->use_ohci_by_default
) {
1970 pci_create_simple(phb
->bus
, -1, "pci-ohci");
1972 pci_create_simple(phb
->bus
, -1, "nec-usb-xhci");
1975 if (spapr
->has_graphics
) {
1976 USBBus
*usb_bus
= usb_bus_find(-1);
1978 usb_create_simple(usb_bus
, "usb-kbd");
1979 usb_create_simple(usb_bus
, "usb-mouse");
1983 if (spapr
->rma_size
< (MIN_RMA_SLOF
<< 20)) {
1985 "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
1990 if (kernel_filename
) {
1991 uint64_t lowaddr
= 0;
1993 kernel_size
= load_elf(kernel_filename
, translate_kernel_address
, NULL
,
1994 NULL
, &lowaddr
, NULL
, 1, PPC_ELF_MACHINE
,
1996 if (kernel_size
== ELF_LOAD_WRONG_ENDIAN
) {
1997 kernel_size
= load_elf(kernel_filename
,
1998 translate_kernel_address
, NULL
,
1999 NULL
, &lowaddr
, NULL
, 0, PPC_ELF_MACHINE
,
2001 kernel_le
= kernel_size
> 0;
2003 if (kernel_size
< 0) {
2004 error_report("error loading %s: %s",
2005 kernel_filename
, load_elf_strerror(kernel_size
));
2010 if (initrd_filename
) {
2011 /* Try to locate the initrd in the gap between the kernel
2012 * and the firmware. Add a bit of space just in case
2014 initrd_base
= (KERNEL_LOAD_ADDR
+ kernel_size
+ 0x1ffff) & ~0xffff;
2015 initrd_size
= load_image_targphys(initrd_filename
, initrd_base
,
2016 load_limit
- initrd_base
);
2017 if (initrd_size
< 0) {
2018 error_report("could not load initial ram disk '%s'",
2028 if (bios_name
== NULL
) {
2029 bios_name
= FW_FILE_NAME
;
2031 filename
= qemu_find_file(QEMU_FILE_TYPE_BIOS
, bios_name
);
2033 error_report("Could not find LPAR firmware '%s'", bios_name
);
2036 fw_size
= load_image_targphys(filename
, 0, FW_MAX_SIZE
);
2038 error_report("Could not load LPAR firmware '%s'", filename
);
2043 /* FIXME: Should register things through the MachineState's qdev
2044 * interface, this is a legacy from the sPAPREnvironment structure
2045 * which predated MachineState but had a similar function */
2046 vmstate_register(NULL
, 0, &vmstate_spapr
, spapr
);
2047 register_savevm_live(NULL
, "spapr/htab", -1, 1,
2048 &savevm_htab_handlers
, spapr
);
2050 /* Prepare the device tree */
2051 spapr
->fdt_skel
= spapr_create_fdt_skel(initrd_base
, initrd_size
,
2052 kernel_size
, kernel_le
,
2054 spapr
->check_exception_irq
);
2055 assert(spapr
->fdt_skel
!= NULL
);
2058 QTAILQ_INIT(&spapr
->ccs_list
);
2059 qemu_register_reset(spapr_ccs_reset_hook
, spapr
);
2061 qemu_register_boot_set(spapr_boot_set
, spapr
);
2064 static int spapr_kvm_type(const char *vm_type
)
2070 if (!strcmp(vm_type
, "HV")) {
2074 if (!strcmp(vm_type
, "PR")) {
2078 error_report("Unknown kvm-type specified '%s'", vm_type
);
2083 * Implementation of an interface to adjust firmware path
2084 * for the bootindex property handling.
2086 static char *spapr_get_fw_dev_path(FWPathProvider
*p
, BusState
*bus
,
2089 #define CAST(type, obj, name) \
2090 ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2091 SCSIDevice
*d
= CAST(SCSIDevice
, dev
, TYPE_SCSI_DEVICE
);
2092 sPAPRPHBState
*phb
= CAST(sPAPRPHBState
, dev
, TYPE_SPAPR_PCI_HOST_BRIDGE
);
2095 void *spapr
= CAST(void, bus
->parent
, "spapr-vscsi");
2096 VirtIOSCSI
*virtio
= CAST(VirtIOSCSI
, bus
->parent
, TYPE_VIRTIO_SCSI
);
2097 USBDevice
*usb
= CAST(USBDevice
, bus
->parent
, TYPE_USB_DEVICE
);
2101 * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2102 * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2103 * in the top 16 bits of the 64-bit LUN
2105 unsigned id
= 0x8000 | (d
->id
<< 8) | d
->lun
;
2106 return g_strdup_printf("%s@%"PRIX64
, qdev_fw_name(dev
),
2107 (uint64_t)id
<< 48);
2108 } else if (virtio
) {
2110 * We use SRP luns of the form 01000000 | (target << 8) | lun
2111 * in the top 32 bits of the 64-bit LUN
2112 * Note: the quote above is from SLOF and it is wrong,
2113 * the actual binding is:
2114 * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2116 unsigned id
= 0x1000000 | (d
->id
<< 16) | d
->lun
;
2117 return g_strdup_printf("%s@%"PRIX64
, qdev_fw_name(dev
),
2118 (uint64_t)id
<< 32);
2121 * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2122 * in the top 32 bits of the 64-bit LUN
2124 unsigned usb_port
= atoi(usb
->port
->path
);
2125 unsigned id
= 0x1000000 | (usb_port
<< 16) | d
->lun
;
2126 return g_strdup_printf("%s@%"PRIX64
, qdev_fw_name(dev
),
2127 (uint64_t)id
<< 32);
2132 /* Replace "pci" with "pci@800000020000000" */
2133 return g_strdup_printf("pci@%"PRIX64
, phb
->buid
);
2139 static char *spapr_get_kvm_type(Object
*obj
, Error
**errp
)
2141 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
2143 return g_strdup(spapr
->kvm_type
);
2146 static void spapr_set_kvm_type(Object
*obj
, const char *value
, Error
**errp
)
2148 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
2150 g_free(spapr
->kvm_type
);
2151 spapr
->kvm_type
= g_strdup(value
);
2154 static void spapr_machine_initfn(Object
*obj
)
2156 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
2158 spapr
->htab_fd
= -1;
2159 object_property_add_str(obj
, "kvm-type",
2160 spapr_get_kvm_type
, spapr_set_kvm_type
, NULL
);
2161 object_property_set_description(obj
, "kvm-type",
2162 "Specifies the KVM virtualization mode (HV, PR)",
2166 static void spapr_machine_finalizefn(Object
*obj
)
2168 sPAPRMachineState
*spapr
= SPAPR_MACHINE(obj
);
2170 g_free(spapr
->kvm_type
);
2173 static void ppc_cpu_do_nmi_on_cpu(void *arg
)
2177 cpu_synchronize_state(cs
);
2178 ppc_cpu_do_system_reset(cs
);
2181 static void spapr_nmi(NMIState
*n
, int cpu_index
, Error
**errp
)
2186 async_run_on_cpu(cs
, ppc_cpu_do_nmi_on_cpu
, cs
);
2190 static void spapr_add_lmbs(DeviceState
*dev
, uint64_t addr
, uint64_t size
,
2191 uint32_t node
, Error
**errp
)
2193 sPAPRDRConnector
*drc
;
2194 sPAPRDRConnectorClass
*drck
;
2195 uint32_t nr_lmbs
= size
/SPAPR_MEMORY_BLOCK_SIZE
;
2196 int i
, fdt_offset
, fdt_size
;
2199 for (i
= 0; i
< nr_lmbs
; i
++) {
2200 drc
= spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB
,
2201 addr
/SPAPR_MEMORY_BLOCK_SIZE
);
2204 fdt
= create_device_tree(&fdt_size
);
2205 fdt_offset
= spapr_populate_memory_node(fdt
, node
, addr
,
2206 SPAPR_MEMORY_BLOCK_SIZE
);
2208 drck
= SPAPR_DR_CONNECTOR_GET_CLASS(drc
);
2209 drck
->attach(drc
, dev
, fdt
, fdt_offset
, !dev
->hotplugged
, errp
);
2210 addr
+= SPAPR_MEMORY_BLOCK_SIZE
;
2212 /* send hotplug notification to the
2213 * guest only in case of hotplugged memory
2215 if (dev
->hotplugged
) {
2216 spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB
, nr_lmbs
);
2220 static void spapr_memory_plug(HotplugHandler
*hotplug_dev
, DeviceState
*dev
,
2221 uint32_t node
, Error
**errp
)
2223 Error
*local_err
= NULL
;
2224 sPAPRMachineState
*ms
= SPAPR_MACHINE(hotplug_dev
);
2225 PCDIMMDevice
*dimm
= PC_DIMM(dev
);
2226 PCDIMMDeviceClass
*ddc
= PC_DIMM_GET_CLASS(dimm
);
2227 MemoryRegion
*mr
= ddc
->get_memory_region(dimm
);
2228 uint64_t align
= memory_region_get_alignment(mr
);
2229 uint64_t size
= memory_region_size(mr
);
2232 if (size
% SPAPR_MEMORY_BLOCK_SIZE
) {
2233 error_setg(&local_err
, "Hotplugged memory size must be a multiple of "
2234 "%lld MB", SPAPR_MEMORY_BLOCK_SIZE
/M_BYTE
);
2238 pc_dimm_memory_plug(dev
, &ms
->hotplug_memory
, mr
, align
, &local_err
);
2243 addr
= object_property_get_int(OBJECT(dimm
), PC_DIMM_ADDR_PROP
, &local_err
);
2245 pc_dimm_memory_unplug(dev
, &ms
->hotplug_memory
, mr
);
2249 spapr_add_lmbs(dev
, addr
, size
, node
, &error_abort
);
2252 error_propagate(errp
, local_err
);
2255 void *spapr_populate_hotplug_cpu_dt(CPUState
*cs
, int *fdt_offset
,
2256 sPAPRMachineState
*spapr
)
2258 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
2259 DeviceClass
*dc
= DEVICE_GET_CLASS(cs
);
2260 int id
= ppc_get_vcpu_dt_id(cpu
);
2262 int offset
, fdt_size
;
2265 fdt
= create_device_tree(&fdt_size
);
2266 nodename
= g_strdup_printf("%s@%x", dc
->fw_name
, id
);
2267 offset
= fdt_add_subnode(fdt
, 0, nodename
);
2269 spapr_populate_cpu_dt(cs
, fdt
, offset
, spapr
);
2272 *fdt_offset
= offset
;
2276 static void spapr_machine_device_plug(HotplugHandler
*hotplug_dev
,
2277 DeviceState
*dev
, Error
**errp
)
2279 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2281 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
)) {
2284 if (!smc
->dr_lmb_enabled
) {
2285 error_setg(errp
, "Memory hotplug not supported for this machine");
2288 node
= object_property_get_int(OBJECT(dev
), PC_DIMM_NODE_PROP
, errp
);
2292 if (node
< 0 || node
>= MAX_NODES
) {
2293 error_setg(errp
, "Invaild node %d", node
);
2298 * Currently PowerPC kernel doesn't allow hot-adding memory to
2299 * memory-less node, but instead will silently add the memory
2300 * to the first node that has some memory. This causes two
2301 * unexpected behaviours for the user.
2303 * - Memory gets hotplugged to a different node than what the user
2305 * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2306 * to memory-less node, a reboot will set things accordingly
2307 * and the previously hotplugged memory now ends in the right node.
2308 * This appears as if some memory moved from one node to another.
2310 * So until kernel starts supporting memory hotplug to memory-less
2311 * nodes, just prevent such attempts upfront in QEMU.
2313 if (nb_numa_nodes
&& !numa_info
[node
].node_mem
) {
2314 error_setg(errp
, "Can't hotplug memory to memory-less node %d",
2319 spapr_memory_plug(hotplug_dev
, dev
, node
, errp
);
2320 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
2321 spapr_core_plug(hotplug_dev
, dev
, errp
);
2325 static void spapr_machine_device_unplug(HotplugHandler
*hotplug_dev
,
2326 DeviceState
*dev
, Error
**errp
)
2328 sPAPRMachineClass
*smc
= SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2330 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
)) {
2331 error_setg(errp
, "Memory hot unplug not supported by sPAPR");
2332 } else if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
2333 if (!smc
->dr_cpu_enabled
) {
2334 error_setg(errp
, "CPU hot unplug not supported on this machine");
2337 spapr_core_unplug(hotplug_dev
, dev
, errp
);
2341 static void spapr_machine_device_pre_plug(HotplugHandler
*hotplug_dev
,
2342 DeviceState
*dev
, Error
**errp
)
2344 if (object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
2345 spapr_core_pre_plug(hotplug_dev
, dev
, errp
);
2349 static HotplugHandler
*spapr_get_hotpug_handler(MachineState
*machine
,
2352 if (object_dynamic_cast(OBJECT(dev
), TYPE_PC_DIMM
) ||
2353 object_dynamic_cast(OBJECT(dev
), TYPE_SPAPR_CPU_CORE
)) {
2354 return HOTPLUG_HANDLER(machine
);
2359 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index
)
2361 /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2362 * socket means much for the paravirtualized PAPR platform) */
2363 return cpu_index
/ smp_threads
/ smp_cores
;
2366 static HotpluggableCPUList
*spapr_query_hotpluggable_cpus(MachineState
*machine
)
2369 HotpluggableCPUList
*head
= NULL
;
2370 sPAPRMachineState
*spapr
= SPAPR_MACHINE(machine
);
2371 int spapr_max_cores
= max_cpus
/ smp_threads
;
2372 int smt
= kvmppc_smt_threads();
2374 for (i
= 0; i
< spapr_max_cores
; i
++) {
2375 HotpluggableCPUList
*list_item
= g_new0(typeof(*list_item
), 1);
2376 HotpluggableCPU
*cpu_item
= g_new0(typeof(*cpu_item
), 1);
2377 CpuInstanceProperties
*cpu_props
= g_new0(typeof(*cpu_props
), 1);
2379 cpu_item
->type
= spapr_get_cpu_core_type(machine
->cpu_model
);
2380 cpu_item
->vcpus_count
= smp_threads
;
2381 cpu_props
->has_core_id
= true;
2382 cpu_props
->core_id
= i
* smt
;
2383 /* TODO: add 'has_node/node' here to describe
2384 to which node core belongs */
2386 cpu_item
->props
= cpu_props
;
2387 if (spapr
->cores
[i
]) {
2388 cpu_item
->has_qom_path
= true;
2389 cpu_item
->qom_path
= object_get_canonical_path(spapr
->cores
[i
]);
2391 list_item
->value
= cpu_item
;
2392 list_item
->next
= head
;
2398 static void spapr_machine_class_init(ObjectClass
*oc
, void *data
)
2400 MachineClass
*mc
= MACHINE_CLASS(oc
);
2401 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(oc
);
2402 FWPathProviderClass
*fwc
= FW_PATH_PROVIDER_CLASS(oc
);
2403 NMIClass
*nc
= NMI_CLASS(oc
);
2404 HotplugHandlerClass
*hc
= HOTPLUG_HANDLER_CLASS(oc
);
2406 mc
->desc
= "pSeries Logical Partition (PAPR compliant)";
2409 * We set up the default / latest behaviour here. The class_init
2410 * functions for the specific versioned machine types can override
2411 * these details for backwards compatibility
2413 mc
->init
= ppc_spapr_init
;
2414 mc
->reset
= ppc_spapr_reset
;
2415 mc
->block_default_type
= IF_SCSI
;
2416 mc
->max_cpus
= MAX_CPUMASK_BITS
;
2417 mc
->no_parallel
= 1;
2418 mc
->default_boot_order
= "";
2419 mc
->default_ram_size
= 512 * M_BYTE
;
2420 mc
->kvm_type
= spapr_kvm_type
;
2421 mc
->has_dynamic_sysbus
= true;
2422 mc
->pci_allow_0_address
= true;
2423 mc
->get_hotplug_handler
= spapr_get_hotpug_handler
;
2424 hc
->pre_plug
= spapr_machine_device_pre_plug
;
2425 hc
->plug
= spapr_machine_device_plug
;
2426 hc
->unplug
= spapr_machine_device_unplug
;
2427 mc
->cpu_index_to_socket_id
= spapr_cpu_index_to_socket_id
;
2428 mc
->query_hotpluggable_cpus
= spapr_query_hotpluggable_cpus
;
2430 smc
->dr_lmb_enabled
= true;
2431 smc
->dr_cpu_enabled
= true;
2432 fwc
->get_dev_path
= spapr_get_fw_dev_path
;
2433 nc
->nmi_monitor_handler
= spapr_nmi
;
2436 static const TypeInfo spapr_machine_info
= {
2437 .name
= TYPE_SPAPR_MACHINE
,
2438 .parent
= TYPE_MACHINE
,
2440 .instance_size
= sizeof(sPAPRMachineState
),
2441 .instance_init
= spapr_machine_initfn
,
2442 .instance_finalize
= spapr_machine_finalizefn
,
2443 .class_size
= sizeof(sPAPRMachineClass
),
2444 .class_init
= spapr_machine_class_init
,
2445 .interfaces
= (InterfaceInfo
[]) {
2446 { TYPE_FW_PATH_PROVIDER
},
2448 { TYPE_HOTPLUG_HANDLER
},
2453 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest) \
2454 static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
2457 MachineClass *mc = MACHINE_CLASS(oc); \
2458 spapr_machine_##suffix##_class_options(mc); \
2460 mc->alias = "pseries"; \
2461 mc->is_default = 1; \
2464 static void spapr_machine_##suffix##_instance_init(Object *obj) \
2466 MachineState *machine = MACHINE(obj); \
2467 spapr_machine_##suffix##_instance_options(machine); \
2469 static const TypeInfo spapr_machine_##suffix##_info = { \
2470 .name = MACHINE_TYPE_NAME("pseries-" verstr), \
2471 .parent = TYPE_SPAPR_MACHINE, \
2472 .class_init = spapr_machine_##suffix##_class_init, \
2473 .instance_init = spapr_machine_##suffix##_instance_init, \
2475 static void spapr_machine_register_##suffix(void) \
2477 type_register(&spapr_machine_##suffix##_info); \
2479 type_init(spapr_machine_register_##suffix)
2484 static void spapr_machine_2_7_instance_options(MachineState
*machine
)
2488 static void spapr_machine_2_7_class_options(MachineClass
*mc
)
2490 /* Defaults for the latest behaviour inherited from the base class */
2493 DEFINE_SPAPR_MACHINE(2_7
, "2.7", true);
2498 #define SPAPR_COMPAT_2_6 \
2501 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,\
2503 .value = stringify(off),\
2506 static void spapr_machine_2_6_instance_options(MachineState
*machine
)
2510 static void spapr_machine_2_6_class_options(MachineClass
*mc
)
2512 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
2514 spapr_machine_2_7_class_options(mc
);
2515 smc
->dr_cpu_enabled
= false;
2516 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_6
);
2519 DEFINE_SPAPR_MACHINE(2_6
, "2.6", false);
2524 #define SPAPR_COMPAT_2_5 \
2527 .driver = "spapr-vlan", \
2528 .property = "use-rx-buffer-pools", \
2532 static void spapr_machine_2_5_instance_options(MachineState
*machine
)
2536 static void spapr_machine_2_5_class_options(MachineClass
*mc
)
2538 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
2540 spapr_machine_2_6_class_options(mc
);
2541 smc
->use_ohci_by_default
= true;
2542 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_5
);
2545 DEFINE_SPAPR_MACHINE(2_5
, "2.5", false);
2550 #define SPAPR_COMPAT_2_4 \
2553 static void spapr_machine_2_4_instance_options(MachineState
*machine
)
2555 spapr_machine_2_5_instance_options(machine
);
2558 static void spapr_machine_2_4_class_options(MachineClass
*mc
)
2560 sPAPRMachineClass
*smc
= SPAPR_MACHINE_CLASS(mc
);
2562 spapr_machine_2_5_class_options(mc
);
2563 smc
->dr_lmb_enabled
= false;
2564 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_4
);
2567 DEFINE_SPAPR_MACHINE(2_4
, "2.4", false);
2572 #define SPAPR_COMPAT_2_3 \
2575 .driver = "spapr-pci-host-bridge",\
2576 .property = "dynamic-reconfiguration",\
2580 static void spapr_machine_2_3_instance_options(MachineState
*machine
)
2582 spapr_machine_2_4_instance_options(machine
);
2583 savevm_skip_section_footers();
2584 global_state_set_optional();
2585 savevm_skip_configuration();
2588 static void spapr_machine_2_3_class_options(MachineClass
*mc
)
2590 spapr_machine_2_4_class_options(mc
);
2591 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_3
);
2593 DEFINE_SPAPR_MACHINE(2_3
, "2.3", false);
2599 #define SPAPR_COMPAT_2_2 \
2602 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,\
2603 .property = "mem_win_size",\
2604 .value = "0x20000000",\
2607 static void spapr_machine_2_2_instance_options(MachineState
*machine
)
2609 spapr_machine_2_3_instance_options(machine
);
2610 machine
->suppress_vmdesc
= true;
2613 static void spapr_machine_2_2_class_options(MachineClass
*mc
)
2615 spapr_machine_2_3_class_options(mc
);
2616 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_2
);
2618 DEFINE_SPAPR_MACHINE(2_2
, "2.2", false);
2623 #define SPAPR_COMPAT_2_1 \
2626 static void spapr_machine_2_1_instance_options(MachineState
*machine
)
2628 spapr_machine_2_2_instance_options(machine
);
2631 static void spapr_machine_2_1_class_options(MachineClass
*mc
)
2633 spapr_machine_2_2_class_options(mc
);
2634 SET_MACHINE_COMPAT(mc
, SPAPR_COMPAT_2_1
);
2636 DEFINE_SPAPR_MACHINE(2_1
, "2.1", false);
2638 static void spapr_machine_register_types(void)
2640 type_register_static(&spapr_machine_info
);
2643 type_init(spapr_machine_register_types
)