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 "sysemu/sysemu.h"
31 #include "sysemu/blockdev.h"
32 #include "sysemu/cpus.h"
33 #include "sysemu/kvm.h"
35 #include "mmu-hash64.h"
37 #include "hw/boards.h"
38 #include "hw/ppc/ppc.h"
39 #include "hw/loader.h"
41 #include "hw/ppc/spapr.h"
42 #include "hw/ppc/spapr_vio.h"
43 #include "hw/pci-host/spapr.h"
44 #include "hw/ppc/xics.h"
45 #include "hw/pci/msi.h"
47 #include "hw/pci/pci.h"
49 #include "exec/address-spaces.h"
51 #include "qemu/config-file.h"
52 #include "qemu/error-report.h"
56 /* SLOF memory layout:
58 * SLOF raw image loaded at 0, copies its romfs right below the flat
59 * device-tree, then position SLOF itself 31M below that
61 * So we set FW_OVERHEAD to 40MB which should account for all of that
64 * We load our kernel at 4M, leaving space for SLOF initial image
66 #define FDT_MAX_SIZE 0x40000
67 #define RTAS_MAX_SIZE 0x10000
68 #define FW_MAX_SIZE 0x400000
69 #define FW_FILE_NAME "slof.bin"
70 #define FW_OVERHEAD 0x2800000
71 #define KERNEL_LOAD_ADDR FW_MAX_SIZE
73 #define MIN_RMA_SLOF 128UL
75 #define TIMEBASE_FREQ 512000000ULL
78 #define XICS_IRQS 1024
80 #define PHANDLE_XICP 0x00001111
82 #define HTAB_SIZE(spapr) (1ULL << ((spapr)->htab_shift))
84 sPAPREnvironment
*spapr
;
86 int spapr_allocate_irq(int hint
, bool lsi
)
92 if (hint
>= spapr
->next_irq
) {
93 spapr
->next_irq
= hint
+ 1;
95 /* FIXME: we should probably check for collisions somehow */
97 irq
= spapr
->next_irq
++;
100 /* Configure irq type */
101 if (!xics_get_qirq(spapr
->icp
, irq
)) {
105 xics_set_irq_type(spapr
->icp
, irq
, lsi
);
111 * Allocate block of consequtive IRQs, returns a number of the first.
112 * If msi==true, aligns the first IRQ number to num.
114 int spapr_allocate_irq_block(int num
, bool lsi
, bool msi
)
120 * MSIMesage::data is used for storing VIRQ so
121 * it has to be aligned to num to support multiple
122 * MSI vectors. MSI-X is not affected by this.
123 * The hint is used for the first IRQ, the rest should
124 * be allocated continuously.
127 assert((num
== 1) || (num
== 2) || (num
== 4) ||
128 (num
== 8) || (num
== 16) || (num
== 32));
129 hint
= (spapr
->next_irq
+ num
- 1) & ~(num
- 1);
132 for (i
= 0; i
< num
; ++i
) {
135 irq
= spapr_allocate_irq(hint
, lsi
);
145 /* If the above doesn't create a consecutive block then that's
147 assert(irq
== (first
+ i
));
153 static XICSState
*try_create_xics(const char *type
, int nr_servers
,
158 dev
= qdev_create(NULL
, type
);
159 qdev_prop_set_uint32(dev
, "nr_servers", nr_servers
);
160 qdev_prop_set_uint32(dev
, "nr_irqs", nr_irqs
);
161 if (qdev_init(dev
) < 0) {
165 return XICS_COMMON(dev
);
168 static XICSState
*xics_system_init(int nr_servers
, int nr_irqs
)
170 XICSState
*icp
= NULL
;
173 QemuOpts
*machine_opts
= qemu_get_machine_opts();
174 bool irqchip_allowed
= qemu_opt_get_bool(machine_opts
,
175 "kernel_irqchip", true);
176 bool irqchip_required
= qemu_opt_get_bool(machine_opts
,
177 "kernel_irqchip", false);
178 if (irqchip_allowed
) {
179 icp
= try_create_xics(TYPE_KVM_XICS
, nr_servers
, nr_irqs
);
182 if (irqchip_required
&& !icp
) {
183 perror("Failed to create in-kernel XICS\n");
189 icp
= try_create_xics(TYPE_XICS
, nr_servers
, nr_irqs
);
193 perror("Failed to create XICS\n");
200 static int spapr_fixup_cpu_dt(void *fdt
, sPAPREnvironment
*spapr
)
205 int smt
= kvmppc_smt_threads();
206 uint32_t pft_size_prop
[] = {0, cpu_to_be32(spapr
->htab_shift
)};
209 DeviceClass
*dc
= DEVICE_GET_CLASS(cpu
);
210 int index
= ppc_get_vcpu_dt_id(POWERPC_CPU(cpu
));
211 uint32_t associativity
[] = {cpu_to_be32(0x5),
215 cpu_to_be32(cpu
->numa_node
),
218 if ((index
% smt
) != 0) {
222 snprintf(cpu_model
, 32, "/cpus/%s@%x", dc
->fw_name
,
225 offset
= fdt_path_offset(fdt
, cpu_model
);
230 if (nb_numa_nodes
> 1) {
231 ret
= fdt_setprop(fdt
, offset
, "ibm,associativity", associativity
,
232 sizeof(associativity
));
238 ret
= fdt_setprop(fdt
, offset
, "ibm,pft-size",
239 pft_size_prop
, sizeof(pft_size_prop
));
248 static size_t create_page_sizes_prop(CPUPPCState
*env
, uint32_t *prop
,
251 size_t maxcells
= maxsize
/ sizeof(uint32_t);
255 for (i
= 0; i
< PPC_PAGE_SIZES_MAX_SZ
; i
++) {
256 struct ppc_one_seg_page_size
*sps
= &env
->sps
.sps
[i
];
258 if (!sps
->page_shift
) {
261 for (count
= 0; count
< PPC_PAGE_SIZES_MAX_SZ
; count
++) {
262 if (sps
->enc
[count
].page_shift
== 0) {
266 if ((p
- prop
) >= (maxcells
- 3 - count
* 2)) {
269 *(p
++) = cpu_to_be32(sps
->page_shift
);
270 *(p
++) = cpu_to_be32(sps
->slb_enc
);
271 *(p
++) = cpu_to_be32(count
);
272 for (j
= 0; j
< count
; j
++) {
273 *(p
++) = cpu_to_be32(sps
->enc
[j
].page_shift
);
274 *(p
++) = cpu_to_be32(sps
->enc
[j
].pte_enc
);
278 return (p
- prop
) * sizeof(uint32_t);
285 fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
286 #exp, fdt_strerror(ret)); \
292 static void *spapr_create_fdt_skel(hwaddr initrd_base
,
296 const char *boot_device
,
297 const char *kernel_cmdline
,
302 uint32_t start_prop
= cpu_to_be32(initrd_base
);
303 uint32_t end_prop
= cpu_to_be32(initrd_base
+ initrd_size
);
304 char hypertas_prop
[] = "hcall-pft\0hcall-term\0hcall-dabr\0hcall-interrupt"
305 "\0hcall-tce\0hcall-vio\0hcall-splpar\0hcall-bulk\0hcall-set-mode";
306 char qemu_hypertas_prop
[] = "hcall-memop1";
307 uint32_t refpoints
[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
308 uint32_t interrupt_server_ranges_prop
[] = {0, cpu_to_be32(smp_cpus
)};
309 int i
, smt
= kvmppc_smt_threads();
310 unsigned char vec5
[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
312 fdt
= g_malloc0(FDT_MAX_SIZE
);
313 _FDT((fdt_create(fdt
, FDT_MAX_SIZE
)));
316 _FDT((fdt_add_reservemap_entry(fdt
, KERNEL_LOAD_ADDR
, kernel_size
)));
319 _FDT((fdt_add_reservemap_entry(fdt
, initrd_base
, initrd_size
)));
321 _FDT((fdt_finish_reservemap(fdt
)));
324 _FDT((fdt_begin_node(fdt
, "")));
325 _FDT((fdt_property_string(fdt
, "device_type", "chrp")));
326 _FDT((fdt_property_string(fdt
, "model", "IBM pSeries (emulated by qemu)")));
327 _FDT((fdt_property_string(fdt
, "compatible", "qemu,pseries")));
329 _FDT((fdt_property_cell(fdt
, "#address-cells", 0x2)));
330 _FDT((fdt_property_cell(fdt
, "#size-cells", 0x2)));
333 _FDT((fdt_begin_node(fdt
, "chosen")));
335 /* Set Form1_affinity */
336 _FDT((fdt_property(fdt
, "ibm,architecture-vec-5", vec5
, sizeof(vec5
))));
338 _FDT((fdt_property_string(fdt
, "bootargs", kernel_cmdline
)));
339 _FDT((fdt_property(fdt
, "linux,initrd-start",
340 &start_prop
, sizeof(start_prop
))));
341 _FDT((fdt_property(fdt
, "linux,initrd-end",
342 &end_prop
, sizeof(end_prop
))));
344 uint64_t kprop
[2] = { cpu_to_be64(KERNEL_LOAD_ADDR
),
345 cpu_to_be64(kernel_size
) };
347 _FDT((fdt_property(fdt
, "qemu,boot-kernel", &kprop
, sizeof(kprop
))));
349 _FDT((fdt_property(fdt
, "qemu,boot-kernel-le", NULL
, 0)));
353 _FDT((fdt_property_string(fdt
, "qemu,boot-device", boot_device
)));
355 _FDT((fdt_property_cell(fdt
, "qemu,graphic-width", graphic_width
)));
356 _FDT((fdt_property_cell(fdt
, "qemu,graphic-height", graphic_height
)));
357 _FDT((fdt_property_cell(fdt
, "qemu,graphic-depth", graphic_depth
)));
359 _FDT((fdt_end_node(fdt
)));
362 _FDT((fdt_begin_node(fdt
, "cpus")));
364 _FDT((fdt_property_cell(fdt
, "#address-cells", 0x1)));
365 _FDT((fdt_property_cell(fdt
, "#size-cells", 0x0)));
368 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
369 CPUPPCState
*env
= &cpu
->env
;
370 DeviceClass
*dc
= DEVICE_GET_CLASS(cs
);
371 PowerPCCPUClass
*pcc
= POWERPC_CPU_GET_CLASS(cs
);
372 int index
= ppc_get_vcpu_dt_id(cpu
);
373 uint32_t servers_prop
[smp_threads
];
374 uint32_t gservers_prop
[smp_threads
* 2];
376 uint32_t segs
[] = {cpu_to_be32(28), cpu_to_be32(40),
377 0xffffffff, 0xffffffff};
378 uint32_t tbfreq
= kvm_enabled() ? kvmppc_get_tbfreq() : TIMEBASE_FREQ
;
379 uint32_t cpufreq
= kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
380 uint32_t page_sizes_prop
[64];
381 size_t page_sizes_prop_size
;
383 if ((index
% smt
) != 0) {
387 nodename
= g_strdup_printf("%s@%x", dc
->fw_name
, index
);
389 _FDT((fdt_begin_node(fdt
, nodename
)));
393 _FDT((fdt_property_cell(fdt
, "reg", index
)));
394 _FDT((fdt_property_string(fdt
, "device_type", "cpu")));
396 _FDT((fdt_property_cell(fdt
, "cpu-version", env
->spr
[SPR_PVR
])));
397 _FDT((fdt_property_cell(fdt
, "d-cache-block-size",
398 env
->dcache_line_size
)));
399 _FDT((fdt_property_cell(fdt
, "d-cache-line-size",
400 env
->dcache_line_size
)));
401 _FDT((fdt_property_cell(fdt
, "i-cache-block-size",
402 env
->icache_line_size
)));
403 _FDT((fdt_property_cell(fdt
, "i-cache-line-size",
404 env
->icache_line_size
)));
406 if (pcc
->l1_dcache_size
) {
407 _FDT((fdt_property_cell(fdt
, "d-cache-size", pcc
->l1_dcache_size
)));
409 fprintf(stderr
, "Warning: Unknown L1 dcache size for cpu\n");
411 if (pcc
->l1_icache_size
) {
412 _FDT((fdt_property_cell(fdt
, "i-cache-size", pcc
->l1_icache_size
)));
414 fprintf(stderr
, "Warning: Unknown L1 icache size for cpu\n");
417 _FDT((fdt_property_cell(fdt
, "timebase-frequency", tbfreq
)));
418 _FDT((fdt_property_cell(fdt
, "clock-frequency", cpufreq
)));
419 _FDT((fdt_property_cell(fdt
, "ibm,slb-size", env
->slb_nr
)));
420 _FDT((fdt_property_string(fdt
, "status", "okay")));
421 _FDT((fdt_property(fdt
, "64-bit", NULL
, 0)));
423 /* Build interrupt servers and gservers properties */
424 for (i
= 0; i
< smp_threads
; i
++) {
425 servers_prop
[i
] = cpu_to_be32(index
+ i
);
426 /* Hack, direct the group queues back to cpu 0 */
427 gservers_prop
[i
*2] = cpu_to_be32(index
+ i
);
428 gservers_prop
[i
*2 + 1] = 0;
430 _FDT((fdt_property(fdt
, "ibm,ppc-interrupt-server#s",
431 servers_prop
, sizeof(servers_prop
))));
432 _FDT((fdt_property(fdt
, "ibm,ppc-interrupt-gserver#s",
433 gservers_prop
, sizeof(gservers_prop
))));
435 if (env
->spr_cb
[SPR_PURR
].oea_read
) {
436 _FDT((fdt_property(fdt
, "ibm,purr", NULL
, 0)));
439 if (env
->mmu_model
& POWERPC_MMU_1TSEG
) {
440 _FDT((fdt_property(fdt
, "ibm,processor-segment-sizes",
441 segs
, sizeof(segs
))));
444 /* Advertise VMX/VSX (vector extensions) if available
445 * 0 / no property == no vector extensions
446 * 1 == VMX / Altivec available
447 * 2 == VSX available */
448 if (env
->insns_flags
& PPC_ALTIVEC
) {
449 uint32_t vmx
= (env
->insns_flags2
& PPC2_VSX
) ? 2 : 1;
451 _FDT((fdt_property_cell(fdt
, "ibm,vmx", vmx
)));
454 /* Advertise DFP (Decimal Floating Point) if available
455 * 0 / no property == no DFP
456 * 1 == DFP available */
457 if (env
->insns_flags2
& PPC2_DFP
) {
458 _FDT((fdt_property_cell(fdt
, "ibm,dfp", 1)));
461 page_sizes_prop_size
= create_page_sizes_prop(env
, page_sizes_prop
,
462 sizeof(page_sizes_prop
));
463 if (page_sizes_prop_size
) {
464 _FDT((fdt_property(fdt
, "ibm,segment-page-sizes",
465 page_sizes_prop
, page_sizes_prop_size
)));
468 _FDT((fdt_end_node(fdt
)));
471 _FDT((fdt_end_node(fdt
)));
474 _FDT((fdt_begin_node(fdt
, "rtas")));
476 _FDT((fdt_property(fdt
, "ibm,hypertas-functions", hypertas_prop
,
477 sizeof(hypertas_prop
))));
478 _FDT((fdt_property(fdt
, "qemu,hypertas-functions", qemu_hypertas_prop
,
479 sizeof(qemu_hypertas_prop
))));
481 _FDT((fdt_property(fdt
, "ibm,associativity-reference-points",
482 refpoints
, sizeof(refpoints
))));
484 _FDT((fdt_property_cell(fdt
, "rtas-error-log-max", RTAS_ERROR_LOG_MAX
)));
486 _FDT((fdt_end_node(fdt
)));
488 /* interrupt controller */
489 _FDT((fdt_begin_node(fdt
, "interrupt-controller")));
491 _FDT((fdt_property_string(fdt
, "device_type",
492 "PowerPC-External-Interrupt-Presentation")));
493 _FDT((fdt_property_string(fdt
, "compatible", "IBM,ppc-xicp")));
494 _FDT((fdt_property(fdt
, "interrupt-controller", NULL
, 0)));
495 _FDT((fdt_property(fdt
, "ibm,interrupt-server-ranges",
496 interrupt_server_ranges_prop
,
497 sizeof(interrupt_server_ranges_prop
))));
498 _FDT((fdt_property_cell(fdt
, "#interrupt-cells", 2)));
499 _FDT((fdt_property_cell(fdt
, "linux,phandle", PHANDLE_XICP
)));
500 _FDT((fdt_property_cell(fdt
, "phandle", PHANDLE_XICP
)));
502 _FDT((fdt_end_node(fdt
)));
505 _FDT((fdt_begin_node(fdt
, "vdevice")));
507 _FDT((fdt_property_string(fdt
, "device_type", "vdevice")));
508 _FDT((fdt_property_string(fdt
, "compatible", "IBM,vdevice")));
509 _FDT((fdt_property_cell(fdt
, "#address-cells", 0x1)));
510 _FDT((fdt_property_cell(fdt
, "#size-cells", 0x0)));
511 _FDT((fdt_property_cell(fdt
, "#interrupt-cells", 0x2)));
512 _FDT((fdt_property(fdt
, "interrupt-controller", NULL
, 0)));
514 _FDT((fdt_end_node(fdt
)));
517 spapr_events_fdt_skel(fdt
, epow_irq
);
519 _FDT((fdt_end_node(fdt
))); /* close root node */
520 _FDT((fdt_finish(fdt
)));
525 static int spapr_populate_memory(sPAPREnvironment
*spapr
, void *fdt
)
527 uint32_t associativity
[] = {cpu_to_be32(0x4), cpu_to_be32(0x0),
528 cpu_to_be32(0x0), cpu_to_be32(0x0),
531 hwaddr node0_size
, mem_start
, node_size
;
532 uint64_t mem_reg_property
[2];
536 if (nb_numa_nodes
> 1 && node_mem
[0] < ram_size
) {
537 node0_size
= node_mem
[0];
539 node0_size
= ram_size
;
543 mem_reg_property
[0] = 0;
544 mem_reg_property
[1] = cpu_to_be64(spapr
->rma_size
);
545 off
= fdt_add_subnode(fdt
, 0, "memory@0");
547 _FDT((fdt_setprop_string(fdt
, off
, "device_type", "memory")));
548 _FDT((fdt_setprop(fdt
, off
, "reg", mem_reg_property
,
549 sizeof(mem_reg_property
))));
550 _FDT((fdt_setprop(fdt
, off
, "ibm,associativity", associativity
,
551 sizeof(associativity
))));
554 if (node0_size
> spapr
->rma_size
) {
555 mem_reg_property
[0] = cpu_to_be64(spapr
->rma_size
);
556 mem_reg_property
[1] = cpu_to_be64(node0_size
- spapr
->rma_size
);
558 sprintf(mem_name
, "memory@" TARGET_FMT_lx
, spapr
->rma_size
);
559 off
= fdt_add_subnode(fdt
, 0, mem_name
);
561 _FDT((fdt_setprop_string(fdt
, off
, "device_type", "memory")));
562 _FDT((fdt_setprop(fdt
, off
, "reg", mem_reg_property
,
563 sizeof(mem_reg_property
))));
564 _FDT((fdt_setprop(fdt
, off
, "ibm,associativity", associativity
,
565 sizeof(associativity
))));
568 /* RAM: Node 1 and beyond */
569 mem_start
= node0_size
;
570 for (i
= 1; i
< nb_numa_nodes
; i
++) {
571 mem_reg_property
[0] = cpu_to_be64(mem_start
);
572 if (mem_start
>= ram_size
) {
575 node_size
= node_mem
[i
];
576 if (node_size
> ram_size
- mem_start
) {
577 node_size
= ram_size
- mem_start
;
580 mem_reg_property
[1] = cpu_to_be64(node_size
);
581 associativity
[3] = associativity
[4] = cpu_to_be32(i
);
582 sprintf(mem_name
, "memory@" TARGET_FMT_lx
, mem_start
);
583 off
= fdt_add_subnode(fdt
, 0, mem_name
);
585 _FDT((fdt_setprop_string(fdt
, off
, "device_type", "memory")));
586 _FDT((fdt_setprop(fdt
, off
, "reg", mem_reg_property
,
587 sizeof(mem_reg_property
))));
588 _FDT((fdt_setprop(fdt
, off
, "ibm,associativity", associativity
,
589 sizeof(associativity
))));
590 mem_start
+= node_size
;
596 static void spapr_finalize_fdt(sPAPREnvironment
*spapr
,
605 fdt
= g_malloc(FDT_MAX_SIZE
);
607 /* open out the base tree into a temp buffer for the final tweaks */
608 _FDT((fdt_open_into(spapr
->fdt_skel
, fdt
, FDT_MAX_SIZE
)));
610 ret
= spapr_populate_memory(spapr
, fdt
);
612 fprintf(stderr
, "couldn't setup memory nodes in fdt\n");
616 ret
= spapr_populate_vdevice(spapr
->vio_bus
, fdt
);
618 fprintf(stderr
, "couldn't setup vio devices in fdt\n");
622 QLIST_FOREACH(phb
, &spapr
->phbs
, list
) {
623 ret
= spapr_populate_pci_dt(phb
, PHANDLE_XICP
, fdt
);
627 fprintf(stderr
, "couldn't setup PCI devices in fdt\n");
632 ret
= spapr_rtas_device_tree_setup(fdt
, rtas_addr
, rtas_size
);
634 fprintf(stderr
, "Couldn't set up RTAS device tree properties\n");
637 /* Advertise NUMA via ibm,associativity */
638 ret
= spapr_fixup_cpu_dt(fdt
, spapr
);
640 fprintf(stderr
, "Couldn't finalize CPU device tree properties\n");
643 if (!spapr
->has_graphics
) {
644 spapr_populate_chosen_stdout(fdt
, spapr
->vio_bus
);
647 _FDT((fdt_pack(fdt
)));
649 if (fdt_totalsize(fdt
) > FDT_MAX_SIZE
) {
650 hw_error("FDT too big ! 0x%x bytes (max is 0x%x)\n",
651 fdt_totalsize(fdt
), FDT_MAX_SIZE
);
655 cpu_physical_memory_write(fdt_addr
, fdt
, fdt_totalsize(fdt
));
660 static uint64_t translate_kernel_address(void *opaque
, uint64_t addr
)
662 return (addr
& 0x0fffffff) + KERNEL_LOAD_ADDR
;
665 static void emulate_spapr_hypercall(PowerPCCPU
*cpu
)
667 CPUPPCState
*env
= &cpu
->env
;
670 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
671 env
->gpr
[3] = H_PRIVILEGE
;
673 env
->gpr
[3] = spapr_hypercall(cpu
, env
->gpr
[3], &env
->gpr
[4]);
677 static void spapr_reset_htab(sPAPREnvironment
*spapr
)
681 /* allocate hash page table. For now we always make this 16mb,
682 * later we should probably make it scale to the size of guest
685 shift
= kvmppc_reset_htab(spapr
->htab_shift
);
688 /* Kernel handles htab, we don't need to allocate one */
689 spapr
->htab_shift
= shift
;
690 kvmppc_kern_htab
= true;
693 /* Allocate an htab if we don't yet have one */
694 spapr
->htab
= qemu_memalign(HTAB_SIZE(spapr
), HTAB_SIZE(spapr
));
698 memset(spapr
->htab
, 0, HTAB_SIZE(spapr
));
701 /* Update the RMA size if necessary */
702 if (spapr
->vrma_adjust
) {
703 hwaddr node0_size
= (nb_numa_nodes
> 1) ? node_mem
[0] : ram_size
;
704 spapr
->rma_size
= kvmppc_rma_size(node0_size
, spapr
->htab_shift
);
708 static void ppc_spapr_reset(void)
710 PowerPCCPU
*first_ppc_cpu
;
712 /* Reset the hash table & recalc the RMA */
713 spapr_reset_htab(spapr
);
715 qemu_devices_reset();
718 spapr_finalize_fdt(spapr
, spapr
->fdt_addr
, spapr
->rtas_addr
,
721 /* Set up the entry state */
722 first_ppc_cpu
= POWERPC_CPU(first_cpu
);
723 first_ppc_cpu
->env
.gpr
[3] = spapr
->fdt_addr
;
724 first_ppc_cpu
->env
.gpr
[5] = 0;
725 first_cpu
->halted
= 0;
726 first_ppc_cpu
->env
.nip
= spapr
->entry_point
;
730 static void spapr_cpu_reset(void *opaque
)
732 PowerPCCPU
*cpu
= opaque
;
733 CPUState
*cs
= CPU(cpu
);
734 CPUPPCState
*env
= &cpu
->env
;
738 /* All CPUs start halted. CPU0 is unhalted from the machine level
739 * reset code and the rest are explicitly started up by the guest
740 * using an RTAS call */
743 env
->spr
[SPR_HIOR
] = 0;
745 env
->external_htab
= (uint8_t *)spapr
->htab
;
746 if (kvm_enabled() && !env
->external_htab
) {
748 * HV KVM, set external_htab to 1 so our ppc_hash64_load_hpte*
749 * functions do the right thing.
751 env
->external_htab
= (void *)1;
755 * htab_mask is the mask used to normalize hash value to PTEG index.
756 * htab_shift is log2 of hash table size.
757 * We have 8 hpte per group, and each hpte is 16 bytes.
758 * ie have 128 bytes per hpte entry.
760 env
->htab_mask
= (1ULL << ((spapr
)->htab_shift
- 7)) - 1;
761 env
->spr
[SPR_SDR1
] = (target_ulong
)(uintptr_t)spapr
->htab
|
762 (spapr
->htab_shift
- 18);
765 static void spapr_create_nvram(sPAPREnvironment
*spapr
)
767 DeviceState
*dev
= qdev_create(&spapr
->vio_bus
->bus
, "spapr-nvram");
768 DriveInfo
*dinfo
= drive_get(IF_PFLASH
, 0, 0);
771 qdev_prop_set_drive_nofail(dev
, "drive", dinfo
->bdrv
);
774 qdev_init_nofail(dev
);
776 spapr
->nvram
= (struct sPAPRNVRAM
*)dev
;
779 /* Returns whether we want to use VGA or not */
780 static int spapr_vga_init(PCIBus
*pci_bus
)
782 switch (vga_interface_type
) {
788 return pci_vga_init(pci_bus
) != NULL
;
790 fprintf(stderr
, "This vga model is not supported,"
791 "currently it only supports -vga std\n");
796 static const VMStateDescription vmstate_spapr
= {
799 .minimum_version_id
= 1,
800 .minimum_version_id_old
= 1,
801 .fields
= (VMStateField
[]) {
802 VMSTATE_UINT32(next_irq
, sPAPREnvironment
),
805 VMSTATE_UINT64(rtc_offset
, sPAPREnvironment
),
807 VMSTATE_END_OF_LIST()
811 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
812 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
813 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
814 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
816 static int htab_save_setup(QEMUFile
*f
, void *opaque
)
818 sPAPREnvironment
*spapr
= opaque
;
820 /* "Iteration" header */
821 qemu_put_be32(f
, spapr
->htab_shift
);
824 spapr
->htab_save_index
= 0;
825 spapr
->htab_first_pass
= true;
827 assert(kvm_enabled());
829 spapr
->htab_fd
= kvmppc_get_htab_fd(false);
830 if (spapr
->htab_fd
< 0) {
831 fprintf(stderr
, "Unable to open fd for reading hash table from KVM: %s\n",
841 static void htab_save_first_pass(QEMUFile
*f
, sPAPREnvironment
*spapr
,
844 int htabslots
= HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
;
845 int index
= spapr
->htab_save_index
;
846 int64_t starttime
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
848 assert(spapr
->htab_first_pass
);
853 /* Consume invalid HPTEs */
854 while ((index
< htabslots
)
855 && !HPTE_VALID(HPTE(spapr
->htab
, index
))) {
857 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
860 /* Consume valid HPTEs */
862 while ((index
< htabslots
)
863 && HPTE_VALID(HPTE(spapr
->htab
, index
))) {
865 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
868 if (index
> chunkstart
) {
869 int n_valid
= index
- chunkstart
;
871 qemu_put_be32(f
, chunkstart
);
872 qemu_put_be16(f
, n_valid
);
874 qemu_put_buffer(f
, HPTE(spapr
->htab
, chunkstart
),
875 HASH_PTE_SIZE_64
* n_valid
);
877 if ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - starttime
) > max_ns
) {
881 } while ((index
< htabslots
) && !qemu_file_rate_limit(f
));
883 if (index
>= htabslots
) {
884 assert(index
== htabslots
);
886 spapr
->htab_first_pass
= false;
888 spapr
->htab_save_index
= index
;
891 static int htab_save_later_pass(QEMUFile
*f
, sPAPREnvironment
*spapr
,
894 bool final
= max_ns
< 0;
895 int htabslots
= HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
;
896 int examined
= 0, sent
= 0;
897 int index
= spapr
->htab_save_index
;
898 int64_t starttime
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
900 assert(!spapr
->htab_first_pass
);
903 int chunkstart
, invalidstart
;
905 /* Consume non-dirty HPTEs */
906 while ((index
< htabslots
)
907 && !HPTE_DIRTY(HPTE(spapr
->htab
, index
))) {
913 /* Consume valid dirty HPTEs */
914 while ((index
< htabslots
)
915 && HPTE_DIRTY(HPTE(spapr
->htab
, index
))
916 && HPTE_VALID(HPTE(spapr
->htab
, index
))) {
917 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
922 invalidstart
= index
;
923 /* Consume invalid dirty HPTEs */
924 while ((index
< htabslots
)
925 && HPTE_DIRTY(HPTE(spapr
->htab
, index
))
926 && !HPTE_VALID(HPTE(spapr
->htab
, index
))) {
927 CLEAN_HPTE(HPTE(spapr
->htab
, index
));
932 if (index
> chunkstart
) {
933 int n_valid
= invalidstart
- chunkstart
;
934 int n_invalid
= index
- invalidstart
;
936 qemu_put_be32(f
, chunkstart
);
937 qemu_put_be16(f
, n_valid
);
938 qemu_put_be16(f
, n_invalid
);
939 qemu_put_buffer(f
, HPTE(spapr
->htab
, chunkstart
),
940 HASH_PTE_SIZE_64
* n_valid
);
941 sent
+= index
- chunkstart
;
943 if (!final
&& (qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - starttime
) > max_ns
) {
948 if (examined
>= htabslots
) {
952 if (index
>= htabslots
) {
953 assert(index
== htabslots
);
956 } while ((examined
< htabslots
) && (!qemu_file_rate_limit(f
) || final
));
958 if (index
>= htabslots
) {
959 assert(index
== htabslots
);
963 spapr
->htab_save_index
= index
;
965 return (examined
>= htabslots
) && (sent
== 0) ? 1 : 0;
968 #define MAX_ITERATION_NS 5000000 /* 5 ms */
969 #define MAX_KVM_BUF_SIZE 2048
971 static int htab_save_iterate(QEMUFile
*f
, void *opaque
)
973 sPAPREnvironment
*spapr
= opaque
;
976 /* Iteration header */
980 assert(kvm_enabled());
982 rc
= kvmppc_save_htab(f
, spapr
->htab_fd
,
983 MAX_KVM_BUF_SIZE
, MAX_ITERATION_NS
);
987 } else if (spapr
->htab_first_pass
) {
988 htab_save_first_pass(f
, spapr
, MAX_ITERATION_NS
);
990 rc
= htab_save_later_pass(f
, spapr
, MAX_ITERATION_NS
);
1001 static int htab_save_complete(QEMUFile
*f
, void *opaque
)
1003 sPAPREnvironment
*spapr
= opaque
;
1005 /* Iteration header */
1006 qemu_put_be32(f
, 0);
1011 assert(kvm_enabled());
1013 rc
= kvmppc_save_htab(f
, spapr
->htab_fd
, MAX_KVM_BUF_SIZE
, -1);
1017 close(spapr
->htab_fd
);
1018 spapr
->htab_fd
= -1;
1020 htab_save_later_pass(f
, spapr
, -1);
1024 qemu_put_be32(f
, 0);
1025 qemu_put_be16(f
, 0);
1026 qemu_put_be16(f
, 0);
1031 static int htab_load(QEMUFile
*f
, void *opaque
, int version_id
)
1033 sPAPREnvironment
*spapr
= opaque
;
1034 uint32_t section_hdr
;
1037 if (version_id
< 1 || version_id
> 1) {
1038 fprintf(stderr
, "htab_load() bad version\n");
1042 section_hdr
= qemu_get_be32(f
);
1045 /* First section, just the hash shift */
1046 if (spapr
->htab_shift
!= section_hdr
) {
1053 assert(kvm_enabled());
1055 fd
= kvmppc_get_htab_fd(true);
1057 fprintf(stderr
, "Unable to open fd to restore KVM hash table: %s\n",
1064 uint16_t n_valid
, n_invalid
;
1066 index
= qemu_get_be32(f
);
1067 n_valid
= qemu_get_be16(f
);
1068 n_invalid
= qemu_get_be16(f
);
1070 if ((index
== 0) && (n_valid
== 0) && (n_invalid
== 0)) {
1075 if ((index
+ n_valid
+ n_invalid
) >
1076 (HTAB_SIZE(spapr
) / HASH_PTE_SIZE_64
)) {
1077 /* Bad index in stream */
1078 fprintf(stderr
, "htab_load() bad index %d (%hd+%hd entries) "
1079 "in htab stream (htab_shift=%d)\n", index
, n_valid
, n_invalid
,
1086 qemu_get_buffer(f
, HPTE(spapr
->htab
, index
),
1087 HASH_PTE_SIZE_64
* n_valid
);
1090 memset(HPTE(spapr
->htab
, index
+ n_valid
), 0,
1091 HASH_PTE_SIZE_64
* n_invalid
);
1098 rc
= kvmppc_load_htab_chunk(f
, fd
, index
, n_valid
, n_invalid
);
1113 static SaveVMHandlers savevm_htab_handlers
= {
1114 .save_live_setup
= htab_save_setup
,
1115 .save_live_iterate
= htab_save_iterate
,
1116 .save_live_complete
= htab_save_complete
,
1117 .load_state
= htab_load
,
1120 /* pSeries LPAR / sPAPR hardware init */
1121 static void ppc_spapr_init(QEMUMachineInitArgs
*args
)
1123 ram_addr_t ram_size
= args
->ram_size
;
1124 const char *cpu_model
= args
->cpu_model
;
1125 const char *kernel_filename
= args
->kernel_filename
;
1126 const char *kernel_cmdline
= args
->kernel_cmdline
;
1127 const char *initrd_filename
= args
->initrd_filename
;
1128 const char *boot_device
= args
->boot_order
;
1133 MemoryRegion
*sysmem
= get_system_memory();
1134 MemoryRegion
*ram
= g_new(MemoryRegion
, 1);
1135 hwaddr rma_alloc_size
;
1136 hwaddr node0_size
= (nb_numa_nodes
> 1) ? node_mem
[0] : ram_size
;
1137 uint32_t initrd_base
= 0;
1138 long kernel_size
= 0, initrd_size
= 0;
1139 long load_limit
, rtas_limit
, fw_size
;
1140 bool kernel_le
= false;
1143 msi_supported
= true;
1145 spapr
= g_malloc0(sizeof(*spapr
));
1146 QLIST_INIT(&spapr
->phbs
);
1148 cpu_ppc_hypercall
= emulate_spapr_hypercall
;
1150 /* Allocate RMA if necessary */
1151 rma_alloc_size
= kvmppc_alloc_rma("ppc_spapr.rma", sysmem
);
1153 if (rma_alloc_size
== -1) {
1154 hw_error("qemu: Unable to create RMA\n");
1158 if (rma_alloc_size
&& (rma_alloc_size
< node0_size
)) {
1159 spapr
->rma_size
= rma_alloc_size
;
1161 spapr
->rma_size
= node0_size
;
1163 /* With KVM, we don't actually know whether KVM supports an
1164 * unbounded RMA (PR KVM) or is limited by the hash table size
1165 * (HV KVM using VRMA), so we always assume the latter
1167 * In that case, we also limit the initial allocations for RTAS
1168 * etc... to 256M since we have no way to know what the VRMA size
1169 * is going to be as it depends on the size of the hash table
1170 * isn't determined yet.
1172 if (kvm_enabled()) {
1173 spapr
->vrma_adjust
= 1;
1174 spapr
->rma_size
= MIN(spapr
->rma_size
, 0x10000000);
1178 if (spapr
->rma_size
> node0_size
) {
1179 fprintf(stderr
, "Error: Numa node 0 has to span the RMA (%#08"HWADDR_PRIx
")\n",
1184 /* We place the device tree and RTAS just below either the top of the RMA,
1185 * or just below 2GB, whichever is lowere, so that it can be
1186 * processed with 32-bit real mode code if necessary */
1187 rtas_limit
= MIN(spapr
->rma_size
, 0x80000000);
1188 spapr
->rtas_addr
= rtas_limit
- RTAS_MAX_SIZE
;
1189 spapr
->fdt_addr
= spapr
->rtas_addr
- FDT_MAX_SIZE
;
1190 load_limit
= spapr
->fdt_addr
- FW_OVERHEAD
;
1192 /* We aim for a hash table of size 1/128 the size of RAM. The
1193 * normal rule of thumb is 1/64 the size of RAM, but that's much
1194 * more than needed for the Linux guests we support. */
1195 spapr
->htab_shift
= 18; /* Minimum architected size */
1196 while (spapr
->htab_shift
<= 46) {
1197 if ((1ULL << (spapr
->htab_shift
+ 7)) >= ram_size
) {
1200 spapr
->htab_shift
++;
1203 /* Set up Interrupt Controller before we create the VCPUs */
1204 spapr
->icp
= xics_system_init(smp_cpus
* kvmppc_smt_threads() / smp_threads
,
1206 spapr
->next_irq
= XICS_IRQ_BASE
;
1209 if (cpu_model
== NULL
) {
1210 cpu_model
= kvm_enabled() ? "host" : "POWER7";
1212 for (i
= 0; i
< smp_cpus
; i
++) {
1213 cpu
= cpu_ppc_init(cpu_model
);
1215 fprintf(stderr
, "Unable to find PowerPC CPU definition\n");
1220 /* Set time-base frequency to 512 MHz */
1221 cpu_ppc_tb_init(env
, TIMEBASE_FREQ
);
1223 /* PAPR always has exception vectors in RAM not ROM. To ensure this,
1224 * MSR[IP] should never be set.
1226 env
->msr_mask
&= ~(1 << 6);
1228 /* Tell KVM that we're in PAPR mode */
1229 if (kvm_enabled()) {
1230 kvmppc_set_papr(cpu
);
1233 xics_cpu_setup(spapr
->icp
, cpu
);
1235 qemu_register_reset(spapr_cpu_reset
, cpu
);
1239 spapr
->ram_limit
= ram_size
;
1240 if (spapr
->ram_limit
> rma_alloc_size
) {
1241 ram_addr_t nonrma_base
= rma_alloc_size
;
1242 ram_addr_t nonrma_size
= spapr
->ram_limit
- rma_alloc_size
;
1244 memory_region_init_ram(ram
, NULL
, "ppc_spapr.ram", nonrma_size
);
1245 vmstate_register_ram_global(ram
);
1246 memory_region_add_subregion(sysmem
, nonrma_base
, ram
);
1249 filename
= qemu_find_file(QEMU_FILE_TYPE_BIOS
, "spapr-rtas.bin");
1250 spapr
->rtas_size
= load_image_targphys(filename
, spapr
->rtas_addr
,
1251 rtas_limit
- spapr
->rtas_addr
);
1252 if (spapr
->rtas_size
< 0) {
1253 hw_error("qemu: could not load LPAR rtas '%s'\n", filename
);
1256 if (spapr
->rtas_size
> RTAS_MAX_SIZE
) {
1257 hw_error("RTAS too big ! 0x%lx bytes (max is 0x%x)\n",
1258 spapr
->rtas_size
, RTAS_MAX_SIZE
);
1263 /* Set up EPOW events infrastructure */
1264 spapr_events_init(spapr
);
1266 /* Set up VIO bus */
1267 spapr
->vio_bus
= spapr_vio_bus_init();
1269 for (i
= 0; i
< MAX_SERIAL_PORTS
; i
++) {
1270 if (serial_hds
[i
]) {
1271 spapr_vty_create(spapr
->vio_bus
, serial_hds
[i
]);
1275 /* We always have at least the nvram device on VIO */
1276 spapr_create_nvram(spapr
);
1279 spapr_pci_msi_init(spapr
, SPAPR_PCI_MSI_WINDOW
);
1280 spapr_pci_rtas_init();
1282 phb
= spapr_create_phb(spapr
, 0);
1284 for (i
= 0; i
< nb_nics
; i
++) {
1285 NICInfo
*nd
= &nd_table
[i
];
1288 nd
->model
= g_strdup("ibmveth");
1291 if (strcmp(nd
->model
, "ibmveth") == 0) {
1292 spapr_vlan_create(spapr
->vio_bus
, nd
);
1294 pci_nic_init_nofail(&nd_table
[i
], phb
->bus
, nd
->model
, NULL
);
1298 for (i
= 0; i
<= drive_get_max_bus(IF_SCSI
); i
++) {
1299 spapr_vscsi_create(spapr
->vio_bus
);
1303 if (spapr_vga_init(phb
->bus
)) {
1304 spapr
->has_graphics
= true;
1307 if (usb_enabled(spapr
->has_graphics
)) {
1308 pci_create_simple(phb
->bus
, -1, "pci-ohci");
1309 if (spapr
->has_graphics
) {
1310 usbdevice_create("keyboard");
1311 usbdevice_create("mouse");
1315 if (spapr
->rma_size
< (MIN_RMA_SLOF
<< 20)) {
1316 fprintf(stderr
, "qemu: pSeries SLOF firmware requires >= "
1317 "%ldM guest RMA (Real Mode Area memory)\n", MIN_RMA_SLOF
);
1321 if (kernel_filename
) {
1322 uint64_t lowaddr
= 0;
1324 kernel_size
= load_elf(kernel_filename
, translate_kernel_address
, NULL
,
1325 NULL
, &lowaddr
, NULL
, 1, ELF_MACHINE
, 0);
1326 if (kernel_size
== ELF_LOAD_WRONG_ENDIAN
) {
1327 kernel_size
= load_elf(kernel_filename
,
1328 translate_kernel_address
, NULL
,
1329 NULL
, &lowaddr
, NULL
, 0, ELF_MACHINE
, 0);
1330 kernel_le
= kernel_size
> 0;
1332 if (kernel_size
< 0) {
1333 fprintf(stderr
, "qemu: error loading %s: %s\n",
1334 kernel_filename
, load_elf_strerror(kernel_size
));
1339 if (initrd_filename
) {
1340 /* Try to locate the initrd in the gap between the kernel
1341 * and the firmware. Add a bit of space just in case
1343 initrd_base
= (KERNEL_LOAD_ADDR
+ kernel_size
+ 0x1ffff) & ~0xffff;
1344 initrd_size
= load_image_targphys(initrd_filename
, initrd_base
,
1345 load_limit
- initrd_base
);
1346 if (initrd_size
< 0) {
1347 fprintf(stderr
, "qemu: could not load initial ram disk '%s'\n",
1357 if (bios_name
== NULL
) {
1358 bios_name
= FW_FILE_NAME
;
1360 filename
= qemu_find_file(QEMU_FILE_TYPE_BIOS
, bios_name
);
1361 fw_size
= load_image_targphys(filename
, 0, FW_MAX_SIZE
);
1363 hw_error("qemu: could not load LPAR rtas '%s'\n", filename
);
1368 spapr
->entry_point
= 0x100;
1370 vmstate_register(NULL
, 0, &vmstate_spapr
, spapr
);
1371 register_savevm_live(NULL
, "spapr/htab", -1, 1,
1372 &savevm_htab_handlers
, spapr
);
1374 /* Prepare the device tree */
1375 spapr
->fdt_skel
= spapr_create_fdt_skel(initrd_base
, initrd_size
,
1376 kernel_size
, kernel_le
,
1377 boot_device
, kernel_cmdline
,
1379 assert(spapr
->fdt_skel
!= NULL
);
1382 static int spapr_kvm_type(const char *vm_type
)
1388 if (!strcmp(vm_type
, "HV")) {
1392 if (!strcmp(vm_type
, "PR")) {
1396 error_report("Unknown kvm-type specified '%s'", vm_type
);
1400 static QEMUMachine spapr_machine
= {
1402 .desc
= "pSeries Logical Partition (PAPR compliant)",
1404 .init
= ppc_spapr_init
,
1405 .reset
= ppc_spapr_reset
,
1406 .block_default_type
= IF_SCSI
,
1407 .max_cpus
= MAX_CPUS
,
1409 .default_boot_order
= NULL
,
1410 .kvm_type
= spapr_kvm_type
,
1413 static void spapr_machine_init(void)
1415 qemu_register_machine(&spapr_machine
);
1418 machine_init(spapr_machine_init
);