2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
6 * Copyright (C) 1999,2001-2006 Silicon Graphics, Inc. All rights reserved.
9 #include <linux/module.h>
10 #include <linux/init.h>
11 #include <linux/delay.h>
12 #include <linux/kernel.h>
13 #include <linux/kdev_t.h>
14 #include <linux/string.h>
15 #include <linux/screen_info.h>
16 #include <linux/console.h>
17 #include <linux/timex.h>
18 #include <linux/sched.h>
19 #include <linux/ioport.h>
21 #include <linux/serial.h>
22 #include <linux/irq.h>
23 #include <linux/bootmem.h>
24 #include <linux/mmzone.h>
25 #include <linux/interrupt.h>
26 #include <linux/acpi.h>
27 #include <linux/compiler.h>
28 #include <linux/root_dev.h>
29 #include <linux/nodemask.h>
31 #include <linux/efi.h>
35 #include <asm/machvec.h>
36 #include <asm/system.h>
37 #include <asm/processor.h>
39 #include <asm/sn/arch.h>
40 #include <asm/sn/addrs.h>
41 #include <asm/sn/pda.h>
42 #include <asm/sn/nodepda.h>
43 #include <asm/sn/sn_cpuid.h>
44 #include <asm/sn/simulator.h>
45 #include <asm/sn/leds.h>
46 #include <asm/sn/bte.h>
47 #include <asm/sn/shub_mmr.h>
48 #include <asm/sn/clksupport.h>
49 #include <asm/sn/sn_sal.h>
50 #include <asm/sn/geo.h>
51 #include <asm/sn/sn_feature_sets.h>
52 #include "xtalk/xwidgetdev.h"
53 #include "xtalk/hubdev.h"
54 #include <asm/sn/klconfig.h>
57 DEFINE_PER_CPU(struct pda_s
, pda_percpu
);
59 #define MAX_PHYS_MEMORY (1UL << IA64_MAX_PHYS_BITS) /* Max physical address supported */
61 extern void bte_init_node(nodepda_t
*, cnodeid_t
);
63 extern void sn_timer_init(void);
64 extern unsigned long last_time_offset
;
65 extern void (*ia64_mark_idle
) (int);
66 extern void snidle(int);
68 unsigned long sn_rtc_cycles_per_second
;
69 EXPORT_SYMBOL(sn_rtc_cycles_per_second
);
71 DEFINE_PER_CPU(struct sn_hub_info_s
, __sn_hub_info
);
72 EXPORT_PER_CPU_SYMBOL(__sn_hub_info
);
74 DEFINE_PER_CPU(short, __sn_cnodeid_to_nasid
[MAX_COMPACT_NODES
]);
75 EXPORT_PER_CPU_SYMBOL(__sn_cnodeid_to_nasid
);
77 DEFINE_PER_CPU(struct nodepda_s
*, __sn_nodepda
);
78 EXPORT_PER_CPU_SYMBOL(__sn_nodepda
);
80 char sn_system_serial_number_string
[128];
81 EXPORT_SYMBOL(sn_system_serial_number_string
);
82 u64 sn_partition_serial_number
;
83 EXPORT_SYMBOL(sn_partition_serial_number
);
85 EXPORT_SYMBOL(sn_partition_id
);
87 EXPORT_SYMBOL(sn_system_size
);
88 u8 sn_sharing_domain_size
;
89 EXPORT_SYMBOL(sn_sharing_domain_size
);
91 EXPORT_SYMBOL(sn_coherency_id
);
93 EXPORT_SYMBOL(sn_region_size
);
94 int sn_prom_type
; /* 0=hardware, 1=medusa/realprom, 2=medusa/fakeprom */
96 short physical_node_map
[MAX_NUMALINK_NODES
];
97 static unsigned long sn_prom_features
[MAX_PROM_FEATURE_SETS
];
99 EXPORT_SYMBOL(physical_node_map
);
103 static void sn_init_pdas(char **);
104 static void build_cnode_tables(void);
106 static nodepda_t
*nodepdaindr
[MAX_COMPACT_NODES
];
109 * The format of "screen_info" is strange, and due to early i386-setup
110 * code. This is just enough to make the console code think we're on a
113 struct screen_info sn_screen_info
= {
116 .orig_video_mode
= 3,
117 .orig_video_cols
= 80,
118 .orig_video_ega_bx
= 3,
119 .orig_video_lines
= 25,
120 .orig_video_isVGA
= 1,
121 .orig_video_points
= 16
125 * This routine can only be used during init, since
126 * smp_boot_data is an init data structure.
127 * We have to use smp_boot_data.cpu_phys_id to find
128 * the physical id of the processor because the normal
129 * cpu_physical_id() relies on data structures that
130 * may not be initialized yet.
133 static int __init
pxm_to_nasid(int pxm
)
138 nid
= pxm_to_node(pxm
);
139 for (i
= 0; i
< num_node_memblks
; i
++) {
140 if (node_memblk
[i
].nid
== nid
) {
141 return NASID_GET(node_memblk
[i
].start_paddr
);
148 * early_sn_setup - early setup routine for SN platforms
150 * Sets up an initial console to aid debugging. Intended primarily
151 * for bringup. See start_kernel() in init/main.c.
154 void __init
early_sn_setup(void)
156 efi_system_table_t
*efi_systab
;
157 efi_config_table_t
*config_tables
;
158 struct ia64_sal_systab
*sal_systab
;
159 struct ia64_sal_desc_entry_point
*ep
;
164 * Parse enough of the SAL tables to locate the SAL entry point. Since, console
165 * IO on SN2 is done via SAL calls, early_printk won't work without this.
167 * This code duplicates some of the ACPI table parsing that is in efi.c & sal.c.
168 * Any changes to those file may have to be made here as well.
170 efi_systab
= (efi_system_table_t
*) __va(ia64_boot_param
->efi_systab
);
171 config_tables
= __va(efi_systab
->tables
);
172 for (i
= 0; i
< efi_systab
->nr_tables
; i
++) {
173 if (efi_guidcmp(config_tables
[i
].guid
, SAL_SYSTEM_TABLE_GUID
) ==
175 sal_systab
= __va(config_tables
[i
].table
);
176 p
= (char *)(sal_systab
+ 1);
177 for (j
= 0; j
< sal_systab
->entry_count
; j
++) {
178 if (*p
== SAL_DESC_ENTRY_POINT
) {
179 ep
= (struct ia64_sal_desc_entry_point
181 ia64_sal_handler_init(__va
186 p
+= SAL_DESC_SIZE(*p
);
190 /* Uh-oh, SAL not available?? */
191 printk(KERN_ERR
"failed to find SAL entry point\n");
194 extern int platform_intr_list
[];
195 static int __cpuinitdata shub_1_1_found
;
200 * Set flag for enabling shub specific wars
203 static inline int __cpuinit
is_shub_1_1(int nasid
)
210 id
= REMOTE_HUB_L(nasid
, SH1_SHUB_ID
);
211 rev
= (id
& SH1_SHUB_ID_REVISION_MASK
) >> SH1_SHUB_ID_REVISION_SHFT
;
215 static void __cpuinit
sn_check_for_wars(void)
222 for_each_online_node(cnode
) {
223 if (is_shub_1_1(cnodeid_to_nasid(cnode
)))
230 * Scan the EFI PCDP table (if it exists) for an acceptable VGA console
231 * output device. If one exists, pick it and set sn_legacy_{io,mem} to
232 * reflect the bus offsets needed to address it.
234 * Since pcdp support in SN is not supported in the 2.4 kernel (or at least
235 * the one lbs is based on) just declare the needed structs here.
237 * Reference spec http://www.dig64.org/specifications/DIG64_PCDPv20.pdf
239 * Returns 0 if no acceptable vga is found, !0 otherwise.
241 * Note: This stuff is duped here because Altix requires the PCDP to
242 * locate a usable VGA device due to lack of proper ACPI support. Structures
243 * could be used from drivers/firmware/pcdp.h, but it was decided that moving
244 * this file to a more public location just for Altix use was undesirable.
247 struct hcdp_uart_desc
{
252 u8 signature
[4]; /* should be 'HCDP' */
254 u8 rev
; /* should be >=3 for pcdp, <3 for hcdp */
262 struct hcdp_uart_desc uart
[0]; /* num_type0 of these */
263 /* pcdp descriptors follow */
264 } __attribute__((packed
));
266 struct pcdp_device_desc
{
271 /* interconnect specific structure follows */
272 /* device specific structure follows that */
273 } __attribute__((packed
));
275 struct pcdp_interface_pci
{
276 u8 type
; /* 1 == pci */
290 } __attribute__((packed
));
292 struct pcdp_vga_device
{
294 /* ACPI Extended Address Space Desc follows */
295 } __attribute__((packed
));
297 /* from pcdp_device_desc.primary */
298 #define PCDP_PRIMARY_CONSOLE 0x01
300 /* from pcdp_device_desc.type */
301 #define PCDP_CONSOLE_INOUT 0x0
302 #define PCDP_CONSOLE_DEBUG 0x1
303 #define PCDP_CONSOLE_OUT 0x2
304 #define PCDP_CONSOLE_IN 0x3
305 #define PCDP_CONSOLE_TYPE_VGA 0x8
307 #define PCDP_CONSOLE_VGA (PCDP_CONSOLE_TYPE_VGA | PCDP_CONSOLE_OUT)
309 /* from pcdp_interface_pci.type */
310 #define PCDP_IF_PCI 1
312 /* from pcdp_interface_pci.translation */
313 #define PCDP_PCI_TRANS_IOPORT 0x02
314 #define PCDP_PCI_TRANS_MMIO 0x01
316 #if defined(CONFIG_VT) && defined(CONFIG_VGA_CONSOLE)
322 struct pcdp_device_desc device
;
323 struct pcdp_interface_pci if_pci
;
324 extern struct efi efi
;
326 if (efi
.hcdp
== EFI_INVALID_TABLE_ADDR
)
327 return; /* no hcdp/pcdp table */
329 pcdp
= __va(efi
.hcdp
);
332 return; /* only support PCDP (rev >= 3) */
334 for (bp
= (u8
*)&pcdp
->uart
[pcdp
->num_type0
];
335 bp
< (u8
*)pcdp
+ pcdp
->length
;
336 bp
+= device
.length
) {
337 memcpy(&device
, bp
, sizeof(device
));
338 if (! (device
.primary
& PCDP_PRIMARY_CONSOLE
))
339 continue; /* not primary console */
341 if (device
.type
!= PCDP_CONSOLE_VGA
)
342 continue; /* not VGA descriptor */
344 memcpy(&if_pci
, bp
+sizeof(device
), sizeof(if_pci
));
345 if (if_pci
.type
!= PCDP_IF_PCI
)
346 continue; /* not PCI interconnect */
348 if (if_pci
.translation
& PCDP_PCI_TRANS_IOPORT
)
349 vga_console_iobase
= if_pci
.ioport_tra
;
351 if (if_pci
.translation
& PCDP_PCI_TRANS_MMIO
)
352 vga_console_membase
=
353 if_pci
.mmio_tra
| __IA64_UNCACHED_OFFSET
;
355 break; /* once we find the primary, we're done */
360 static unsigned long sn2_rtc_initial
;
363 * sn_setup - SN platform setup routine
364 * @cmdline_p: kernel command line
366 * Handles platform setup for SN machines. This includes determining
367 * the RTC frequency (via a SAL call), initializing secondary CPUs, and
368 * setting up per-node data areas. The console is also initialized here.
370 void __init
sn_setup(char **cmdline_p
)
372 long status
, ticks_per_sec
, drift
;
373 u32 version
= sn_sal_rev();
374 extern void sn_cpu_init(void);
376 sn2_rtc_initial
= rtc_time();
377 ia64_sn_plat_set_error_handling_features(); // obsolete
378 ia64_sn_set_os_feature(OSF_MCA_SLV_TO_OS_INIT_SLV
);
379 ia64_sn_set_os_feature(OSF_FEAT_LOG_SBES
);
381 * Note: The calls to notify the PROM of ACPI and PCI Segment
382 * support must be done prior to acpi_load_tables(), as
383 * an ACPI capable PROM will rebuild the DSDT as result
386 ia64_sn_set_os_feature(OSF_PCISEGMENT_ENABLE
);
387 ia64_sn_set_os_feature(OSF_ACPI_ENABLE
);
389 /* Load the new DSDT and SSDT tables into the global table list. */
392 #if defined(CONFIG_VT) && defined(CONFIG_VGA_CONSOLE)
394 * Handle SN vga console.
396 * SN systems do not have enough ACPI table information
397 * being passed from prom to identify VGA adapters and the legacy
398 * addresses to access them. Until that is done, SN systems rely
399 * on the PCDP table to identify the primary VGA console if one
402 * However, kernel PCDP support is optional, and even if it is built
403 * into the kernel, it will not be used if the boot cmdline contains
404 * console= directives.
406 * So, to work around this mess, we duplicate some of the PCDP code
407 * here so that the primary VGA console (as defined by PCDP) will
408 * work on SN systems even if a different console (e.g. serial) is
409 * selected on the boot line (or CONFIG_EFI_PCDP is off).
412 if (! vga_console_membase
)
416 * Setup legacy IO space.
417 * vga_console_iobase maps to PCI IO Space address 0 on the
418 * bus containing the VGA console.
420 if (vga_console_iobase
) {
421 io_space
[0].mmio_base
=
422 (unsigned long) ioremap(vga_console_iobase
, 0);
423 io_space
[0].sparse
= 0;
426 if (vga_console_membase
) {
427 /* usable vga ... make tty0 the preferred default console */
428 if (!strstr(*cmdline_p
, "console="))
429 add_preferred_console("tty", 0, NULL
);
431 printk(KERN_DEBUG
"SGI: Disabling VGA console\n");
432 if (!strstr(*cmdline_p
, "console="))
433 add_preferred_console("ttySG", 0, NULL
);
434 #ifdef CONFIG_DUMMY_CONSOLE
435 conswitchp
= &dummy_con
;
438 #endif /* CONFIG_DUMMY_CONSOLE */
440 #endif /* def(CONFIG_VT) && def(CONFIG_VGA_CONSOLE) */
442 MAX_DMA_ADDRESS
= PAGE_OFFSET
+ MAX_PHYS_MEMORY
;
445 * Build the tables for managing cnodes.
447 build_cnode_tables();
450 ia64_sal_freq_base(SAL_FREQ_BASE_REALTIME_CLOCK
, &ticks_per_sec
,
452 if (status
!= 0 || ticks_per_sec
< 100000) {
454 "unable to determine platform RTC clock frequency, guessing.\n");
455 /* PROM gives wrong value for clock freq. so guess */
456 sn_rtc_cycles_per_second
= 1000000000000UL / 30000UL;
458 sn_rtc_cycles_per_second
= ticks_per_sec
;
460 platform_intr_list
[ACPI_INTERRUPT_CPEI
] = IA64_CPE_VECTOR
;
462 printk("SGI SAL version %x.%02x\n", version
>> 8, version
& 0x00FF);
465 * we set the default root device to /dev/hda
466 * to make simulation easy
468 ROOT_DEV
= Root_HDA1
;
471 * Create the PDAs and NODEPDAs for all the cpus.
473 sn_init_pdas(cmdline_p
);
475 ia64_mark_idle
= &snidle
;
478 * For the bootcpu, we do this here. All other cpus will make the
479 * call as part of cpu_init in slave cpu initialization.
486 screen_info
= sn_screen_info
;
491 * set pm_power_off to a SAL call to allow
492 * sn machines to power off. The SAL call can be replaced
493 * by an ACPI interface call when ACPI is fully implemented
496 pm_power_off
= ia64_sn_power_down
;
497 current
->thread
.flags
|= IA64_THREAD_MIGRATION
;
501 * sn_init_pdas - setup node data areas
503 * One time setup for Node Data Area. Called by sn_setup().
505 static void __init
sn_init_pdas(char **cmdline_p
)
510 * Allocate & initialize the nodepda for each node.
512 for_each_online_node(cnode
) {
514 alloc_bootmem_node(NODE_DATA(cnode
), sizeof(nodepda_t
));
515 memset(nodepdaindr
[cnode
]->phys_cpuid
, -1,
516 sizeof(nodepdaindr
[cnode
]->phys_cpuid
));
517 spin_lock_init(&nodepdaindr
[cnode
]->ptc_lock
);
521 * Allocate & initialize nodepda for TIOs. For now, put them on node 0.
523 for (cnode
= num_online_nodes(); cnode
< num_cnodes
; cnode
++)
525 alloc_bootmem_node(NODE_DATA(0), sizeof(nodepda_t
));
528 * Now copy the array of nodepda pointers to each nodepda.
530 for (cnode
= 0; cnode
< num_cnodes
; cnode
++)
531 memcpy(nodepdaindr
[cnode
]->pernode_pdaindr
, nodepdaindr
,
532 sizeof(nodepdaindr
));
535 * Set up IO related platform-dependent nodepda fields.
536 * The following routine actually sets up the hubinfo struct
539 for_each_online_node(cnode
) {
540 bte_init_node(nodepdaindr
[cnode
], cnode
);
544 * Initialize the per node hubdev. This includes IO Nodes and
545 * headless/memless nodes.
547 for (cnode
= 0; cnode
< num_cnodes
; cnode
++) {
548 hubdev_init_node(nodepdaindr
[cnode
], cnode
);
553 * sn_cpu_init - initialize per-cpu data areas
554 * @cpuid: cpuid of the caller
556 * Called during cpu initialization on each cpu as it starts.
557 * Currently, initializes the per-cpu data area for SNIA.
558 * Also sets up a few fields in the nodepda. Also known as
559 * platform_cpu_init() by the ia64 machvec code.
561 void __cpuinit
sn_cpu_init(void)
570 static int wars_have_been_checked
, set_cpu0_number
;
572 cpuid
= smp_processor_id();
573 if (cpuid
== 0 && IS_MEDUSA()) {
574 if (ia64_sn_is_fake_prom())
578 printk(KERN_INFO
"Running on medusa with %s PROM\n",
579 (sn_prom_type
== 1) ? "real" : "fake");
582 memset(pda
, 0, sizeof(pda
));
583 if (ia64_sn_get_sn_info(0, &sn_hub_info
->shub2
,
584 &sn_hub_info
->nasid_bitmask
,
585 &sn_hub_info
->nasid_shift
,
586 &sn_system_size
, &sn_sharing_domain_size
,
587 &sn_partition_id
, &sn_coherency_id
,
590 sn_hub_info
->as_shift
= sn_hub_info
->nasid_shift
- 2;
593 * Don't check status. The SAL call is not supported on all PROMs
594 * but a failure is harmless.
595 * Architecturally, cpu_init is always called twice on cpu 0. We
596 * should set cpu_number on cpu 0 once.
599 if (!set_cpu0_number
) {
600 (void) ia64_sn_set_cpu_number(cpuid
);
604 (void) ia64_sn_set_cpu_number(cpuid
);
607 * The boot cpu makes this call again after platform initialization is
610 if (nodepdaindr
[0] == NULL
)
613 for (i
= 0; i
< MAX_PROM_FEATURE_SETS
; i
++)
614 if (ia64_sn_get_prom_feature_set(i
, &sn_prom_features
[i
]) != 0)
617 cpuphyid
= get_sapicid();
619 if (ia64_sn_get_sapic_info(cpuphyid
, &nasid
, &subnode
, &slice
))
622 for (i
=0; i
< MAX_NUMNODES
; i
++) {
623 if (nodepdaindr
[i
]) {
624 nodepdaindr
[i
]->phys_cpuid
[cpuid
].nasid
= nasid
;
625 nodepdaindr
[i
]->phys_cpuid
[cpuid
].slice
= slice
;
626 nodepdaindr
[i
]->phys_cpuid
[cpuid
].subnode
= subnode
;
630 cnode
= nasid_to_cnodeid(nasid
);
632 sn_nodepda
= nodepdaindr
[cnode
];
635 (typeof(pda
->led_address
)) (LED0
+ (slice
<< LED_CPU_SHIFT
));
636 pda
->led_state
= LED_ALWAYS_SET
;
637 pda
->hb_count
= HZ
/ 2;
642 /* copy cpu 0's sn_cnodeid_to_nasid table to this cpu's */
643 memcpy(sn_cnodeid_to_nasid
,
644 (&per_cpu(__sn_cnodeid_to_nasid
, 0)),
645 sizeof(__ia64_per_cpu_var(__sn_cnodeid_to_nasid
)));
650 * Only needs to be done once, on BSP.
651 * Has to be done after loop above, because it uses this cpu's
652 * sn_cnodeid_to_nasid table which was just initialized if this
654 * Has to be done before assignment below.
656 if (!wars_have_been_checked
) {
658 wars_have_been_checked
= 1;
660 sn_hub_info
->shub_1_1_found
= shub_1_1_found
;
663 * Set up addresses of PIO/MEM write status registers.
666 u64 pio1
[] = {SH1_PIO_WRITE_STATUS_0
, 0, SH1_PIO_WRITE_STATUS_1
, 0};
667 u64 pio2
[] = {SH2_PIO_WRITE_STATUS_0
, SH2_PIO_WRITE_STATUS_2
,
668 SH2_PIO_WRITE_STATUS_1
, SH2_PIO_WRITE_STATUS_3
};
670 pio
= is_shub1() ? pio1
: pio2
;
671 pda
->pio_write_status_addr
=
672 (volatile unsigned long *)GLOBAL_MMR_ADDR(nasid
, pio
[slice
]);
673 pda
->pio_write_status_val
= is_shub1() ? SH_PIO_WRITE_STATUS_PENDING_WRITE_COUNT_MASK
: 0;
677 * WAR addresses for SHUB 1.x.
679 if (local_node_data
->active_cpu_count
++ == 0 && is_shub1()) {
682 cnodeid_to_nasid(numa_node_id() ==
683 num_online_nodes() - 1 ? 0 : numa_node_id() + 1);
684 pda
->pio_shub_war_cam_addr
=
685 (volatile unsigned long *)GLOBAL_MMR_ADDR(nasid
,
691 * Build tables for converting between NASIDs and cnodes.
693 static inline int __init
board_needs_cnode(int type
)
695 return (type
== KLTYPE_SNIA
|| type
== KLTYPE_TIO
);
698 void __init
build_cnode_tables(void)
704 memset(physical_node_map
, -1, sizeof(physical_node_map
));
705 memset(sn_cnodeid_to_nasid
, -1,
706 sizeof(__ia64_per_cpu_var(__sn_cnodeid_to_nasid
)));
709 * First populate the tables with C/M bricks. This ensures that
710 * cnode == node for all C & M bricks.
712 for_each_online_node(node
) {
713 nasid
= pxm_to_nasid(node_to_pxm(node
));
714 sn_cnodeid_to_nasid
[node
] = nasid
;
715 physical_node_map
[nasid
] = node
;
719 * num_cnodes is total number of C/M/TIO bricks. Because of the 256 node
720 * limit on the number of nodes, we can't use the generic node numbers
721 * for this. Note that num_cnodes is incremented below as TIOs or
722 * headless/memoryless nodes are discovered.
724 num_cnodes
= num_online_nodes();
726 /* fakeprom does not support klgraph */
727 if (IS_RUNNING_ON_FAKE_PROM())
730 /* Find TIOs & headless/memoryless nodes and add them to the tables */
731 for_each_online_node(node
) {
732 kl_config_hdr_t
*klgraph_header
;
733 nasid
= cnodeid_to_nasid(node
);
734 klgraph_header
= ia64_sn_get_klconfig_addr(nasid
);
735 BUG_ON(klgraph_header
== NULL
);
736 brd
= NODE_OFFSET_TO_LBOARD(nasid
, klgraph_header
->ch_board_info
);
738 if (board_needs_cnode(brd
->brd_type
) && physical_node_map
[brd
->brd_nasid
] < 0) {
739 sn_cnodeid_to_nasid
[num_cnodes
] = brd
->brd_nasid
;
740 physical_node_map
[brd
->brd_nasid
] = num_cnodes
++;
742 brd
= find_lboard_next(brd
);
748 nasid_slice_to_cpuid(int nasid
, int slice
)
752 for (cpu
= 0; cpu
< nr_cpu_ids
; cpu
++)
753 if (cpuid_to_nasid(cpu
) == nasid
&&
754 cpuid_to_slice(cpu
) == slice
)
760 int sn_prom_feature_available(int id
)
762 if (id
>= BITS_PER_LONG
* MAX_PROM_FEATURE_SETS
)
764 return test_bit(id
, sn_prom_features
);
768 sn_kernel_launch_event(void)
770 /* ignore status until we understand possible failure, if any*/
771 if (ia64_sn_kernel_launch_event())
772 printk(KERN_ERR
"KEXEC is not supported in this PROM, Please update the PROM.\n");
774 EXPORT_SYMBOL(sn_prom_feature_available
);