2 * Intel Multiprocessor Specification 1.1 and 1.4
3 * compliant MP-table parsing routines.
5 * (c) 1995 Alan Cox, Building #3 <alan@redhat.com>
6 * (c) 1998, 1999, 2000 Ingo Molnar <mingo@redhat.com>
7 * (c) 2008 Alexey Starikovskiy <astarikovskiy@suse.de>
11 #include <linux/init.h>
12 #include <linux/delay.h>
13 #include <linux/bootmem.h>
14 #include <linux/kernel_stat.h>
15 #include <linux/mc146818rtc.h>
16 #include <linux/bitops.h>
17 #include <linux/acpi.h>
18 #include <linux/module.h>
22 #include <asm/mpspec.h>
23 #include <asm/pgalloc.h>
24 #include <asm/io_apic.h>
25 #include <asm/proto.h>
27 #include <asm/bios_ebda.h>
29 #include <mach_apic.h>
31 #include <mach_apicdef.h>
32 #include <mach_mpparse.h>
35 /* Have we found an MP table */
39 * Various Linux-internal data structures created from the
42 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
43 int mp_bus_id_to_type[MAX_MP_BUSSES];
46 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
47 int mp_bus_id_to_pci_bus[MAX_MP_BUSSES] = {[0 ... MAX_MP_BUSSES - 1] = -1 };
49 static int mp_current_pci_id;
54 * Intel MP BIOS table parsing routines:
58 * Checksum an MP configuration block.
61 static int __init mpf_checksum(unsigned char *mp, int len)
71 #ifdef CONFIG_X86_NUMAQ
73 * Have to match translation table entries to main table entries by counter
74 * hence the mpc_record variable .... can't see a less disgusting way of
78 static int mpc_record;
79 static struct mpc_config_translation *translation_table[MAX_MPC_ENTRY]
83 static void __cpuinit MP_processor_info(struct mpc_config_processor *m)
86 char *bootup_cpu = "";
88 if (!(m->mpc_cpuflag & CPU_ENABLED)) {
92 #ifdef CONFIG_X86_NUMAQ
93 apicid = mpc_apic_id(m, translation_table[mpc_record]);
95 apicid = m->mpc_apicid;
97 if (m->mpc_cpuflag & CPU_BOOTPROCESSOR) {
98 bootup_cpu = " (Bootup-CPU)";
99 boot_cpu_physical_apicid = m->mpc_apicid;
102 printk(KERN_INFO "Processor #%d%s\n", m->mpc_apicid, bootup_cpu);
103 generic_processor_info(apicid, m->mpc_apicver);
106 static void __init MP_bus_info(struct mpc_config_bus *m)
110 memcpy(str, m->mpc_bustype, 6);
113 #ifdef CONFIG_X86_NUMAQ
114 mpc_oem_bus_info(m, str, translation_table[mpc_record]);
116 printk(KERN_INFO "Bus #%d is %s\n", m->mpc_busid, str);
119 #if MAX_MP_BUSSES < 256
120 if (m->mpc_busid >= MAX_MP_BUSSES) {
121 printk(KERN_WARNING "MP table busid value (%d) for bustype %s "
122 " is too large, max. supported is %d\n",
123 m->mpc_busid, str, MAX_MP_BUSSES - 1);
128 if (strncmp(str, BUSTYPE_ISA, sizeof(BUSTYPE_ISA) - 1) == 0) {
129 set_bit(m->mpc_busid, mp_bus_not_pci);
130 #if defined(CONFIG_EISA) || defined (CONFIG_MCA)
131 mp_bus_id_to_type[m->mpc_busid] = MP_BUS_ISA;
133 } else if (strncmp(str, BUSTYPE_PCI, sizeof(BUSTYPE_PCI) - 1) == 0) {
134 #ifdef CONFIG_X86_NUMAQ
135 mpc_oem_pci_bus(m, translation_table[mpc_record]);
137 clear_bit(m->mpc_busid, mp_bus_not_pci);
138 mp_bus_id_to_pci_bus[m->mpc_busid] = mp_current_pci_id;
140 #if defined(CONFIG_EISA) || defined (CONFIG_MCA)
141 mp_bus_id_to_type[m->mpc_busid] = MP_BUS_PCI;
142 } else if (strncmp(str, BUSTYPE_EISA, sizeof(BUSTYPE_EISA) - 1) == 0) {
143 mp_bus_id_to_type[m->mpc_busid] = MP_BUS_EISA;
144 } else if (strncmp(str, BUSTYPE_MCA, sizeof(BUSTYPE_MCA) - 1) == 0) {
145 mp_bus_id_to_type[m->mpc_busid] = MP_BUS_MCA;
148 printk(KERN_WARNING "Unknown bustype %s - ignoring\n", str);
151 #ifdef CONFIG_X86_IO_APIC
153 static int bad_ioapic(unsigned long address)
155 if (nr_ioapics >= MAX_IO_APICS) {
156 printk(KERN_ERR "ERROR: Max # of I/O APICs (%d) exceeded "
157 "(found %d)\n", MAX_IO_APICS, nr_ioapics);
158 panic("Recompile kernel with bigger MAX_IO_APICS!\n");
161 printk(KERN_ERR "WARNING: Bogus (zero) I/O APIC address"
162 " found in table, skipping!\n");
168 static void __init MP_ioapic_info(struct mpc_config_ioapic *m)
170 if (!(m->mpc_flags & MPC_APIC_USABLE))
173 printk(KERN_INFO "I/O APIC #%d Version %d at 0x%X.\n",
174 m->mpc_apicid, m->mpc_apicver, m->mpc_apicaddr);
176 if (bad_ioapic(m->mpc_apicaddr))
179 mp_ioapics[nr_ioapics].mp_apicaddr = m->mpc_apicaddr;
180 mp_ioapics[nr_ioapics].mp_apicid = m->mpc_apicid;
181 mp_ioapics[nr_ioapics].mp_type = m->mpc_type;
182 mp_ioapics[nr_ioapics].mp_apicver = m->mpc_apicver;
183 mp_ioapics[nr_ioapics].mp_flags = m->mpc_flags;
187 static void __init MP_intsrc_info(struct mpc_config_intsrc *m)
189 mp_irqs[mp_irq_entries] = *m;
190 printk(KERN_INFO "Int: type %d, pol %d, trig %d, bus %02x,"
191 " IRQ %02x, APIC ID %x, APIC INT %02x\n",
192 m->mpc_irqtype, m->mpc_irqflag & 3,
193 (m->mpc_irqflag >> 2) & 3, m->mpc_srcbus,
194 m->mpc_srcbusirq, m->mpc_dstapic, m->mpc_dstirq);
195 if (++mp_irq_entries == MAX_IRQ_SOURCES)
196 panic("Max # of irq sources exceeded!!\n");
201 static void __init MP_lintsrc_info(struct mpc_config_lintsrc *m)
203 printk(KERN_INFO "Lint: type %d, pol %d, trig %d, bus %02x,"
204 " IRQ %02x, APIC ID %x, APIC LINT %02x\n",
205 m->mpc_irqtype, m->mpc_irqflag & 3,
206 (m->mpc_irqflag >> 2) & 3, m->mpc_srcbusid,
207 m->mpc_srcbusirq, m->mpc_destapic, m->mpc_destapiclint);
210 #ifdef CONFIG_X86_NUMAQ
211 static void __init MP_translation_info(struct mpc_config_translation *m)
214 "Translation: record %d, type %d, quad %d, global %d, local %d\n",
215 mpc_record, m->trans_type, m->trans_quad, m->trans_global,
218 if (mpc_record >= MAX_MPC_ENTRY)
219 printk(KERN_ERR "MAX_MPC_ENTRY exceeded!\n");
221 translation_table[mpc_record] = m; /* stash this for later */
222 if (m->trans_quad < MAX_NUMNODES && !node_online(m->trans_quad))
223 node_set_online(m->trans_quad);
227 * Read/parse the MPC oem tables
230 static void __init smp_read_mpc_oem(struct mp_config_oemtable *oemtable,
231 unsigned short oemsize)
233 int count = sizeof(*oemtable); /* the header size */
234 unsigned char *oemptr = ((unsigned char *)oemtable) + count;
237 printk(KERN_INFO "Found an OEM MPC table at %8p - parsing it ... \n",
239 if (memcmp(oemtable->oem_signature, MPC_OEM_SIGNATURE, 4)) {
241 "SMP mpc oemtable: bad signature [%c%c%c%c]!\n",
242 oemtable->oem_signature[0], oemtable->oem_signature[1],
243 oemtable->oem_signature[2], oemtable->oem_signature[3]);
246 if (mpf_checksum((unsigned char *)oemtable, oemtable->oem_length)) {
247 printk(KERN_WARNING "SMP oem mptable: checksum error!\n");
250 while (count < oemtable->oem_length) {
254 struct mpc_config_translation *m =
255 (struct mpc_config_translation *)oemptr;
256 MP_translation_info(m);
257 oemptr += sizeof(*m);
265 "Unrecognised OEM table entry type! - %d\n",
273 static inline void mps_oem_check(struct mp_config_table *mpc, char *oem,
276 if (strncmp(oem, "IBM NUMA", 8))
277 printk("Warning! May not be a NUMA-Q system!\n");
279 smp_read_mpc_oem((struct mp_config_oemtable *)mpc->mpc_oemptr,
282 #endif /* CONFIG_X86_NUMAQ */
288 static int __init smp_read_mpc(struct mp_config_table *mpc, unsigned early)
292 int count = sizeof(*mpc);
293 unsigned char *mpt = ((unsigned char *)mpc) + count;
295 if (memcmp(mpc->mpc_signature, MPC_SIGNATURE, 4)) {
296 printk(KERN_ERR "MPTABLE: bad signature [%c%c%c%c]!\n",
297 mpc->mpc_signature[0], mpc->mpc_signature[1],
298 mpc->mpc_signature[2], mpc->mpc_signature[3]);
301 if (mpf_checksum((unsigned char *)mpc, mpc->mpc_length)) {
302 printk(KERN_ERR "MPTABLE: checksum error!\n");
305 if (mpc->mpc_spec != 0x01 && mpc->mpc_spec != 0x04) {
306 printk(KERN_ERR "MPTABLE: bad table version (%d)!!\n",
310 if (!mpc->mpc_lapic) {
311 printk(KERN_ERR "MPTABLE: null local APIC address!\n");
314 memcpy(oem, mpc->mpc_oem, 8);
316 printk(KERN_INFO "MPTABLE: OEM ID: %s\n", oem);
318 memcpy(str, mpc->mpc_productid, 12);
322 mps_oem_check(mpc, oem, str);
324 printk(KERN_INFO "MPTABLE: Product ID: %s\n", str);
326 printk(KERN_INFO "MPTABLE: APIC at: 0x%X\n", mpc->mpc_lapic);
328 /* save the local APIC address, it might be non-default */
330 mp_lapic_addr = mpc->mpc_lapic;
336 * Now process the configuration blocks.
338 #ifdef CONFIG_X86_NUMAQ
341 while (count < mpc->mpc_length) {
345 struct mpc_config_processor *m =
346 (struct mpc_config_processor *)mpt;
347 /* ACPI may have already provided this data */
349 MP_processor_info(m);
356 struct mpc_config_bus *m =
357 (struct mpc_config_bus *)mpt;
365 #ifdef CONFIG_X86_IO_APIC
366 struct mpc_config_ioapic *m =
367 (struct mpc_config_ioapic *)mpt;
370 mpt += sizeof(struct mpc_config_ioapic);
371 count += sizeof(struct mpc_config_ioapic);
376 #ifdef CONFIG_X86_IO_APIC
377 struct mpc_config_intsrc *m =
378 (struct mpc_config_intsrc *)mpt;
382 mpt += sizeof(struct mpc_config_intsrc);
383 count += sizeof(struct mpc_config_intsrc);
388 struct mpc_config_lintsrc *m =
389 (struct mpc_config_lintsrc *)mpt;
397 printk(KERN_ERR "Your mptable is wrong, contact your HW vendor!\n");
398 printk(KERN_ERR "type %x\n", *mpt);
399 print_hex_dump(KERN_ERR, " ", DUMP_PREFIX_ADDRESS, 16,
400 1, mpc, mpc->mpc_length, 1);
401 count = mpc->mpc_length;
404 #ifdef CONFIG_X86_NUMAQ
408 setup_apic_routing();
410 printk(KERN_ERR "MPTABLE: no processors registered!\n");
411 return num_processors;
414 #ifdef CONFIG_X86_IO_APIC
416 static int __init ELCR_trigger(unsigned int irq)
420 port = 0x4d0 + (irq >> 3);
421 return (inb(port) >> (irq & 7)) & 1;
424 static void __init construct_default_ioirq_mptable(int mpc_default_type)
426 struct mpc_config_intsrc intsrc;
428 int ELCR_fallback = 0;
430 intsrc.mpc_type = MP_INTSRC;
431 intsrc.mpc_irqflag = 0; /* conforming */
432 intsrc.mpc_srcbus = 0;
433 intsrc.mpc_dstapic = mp_ioapics[0].mp_apicid;
435 intsrc.mpc_irqtype = mp_INT;
438 * If true, we have an ISA/PCI system with no IRQ entries
439 * in the MP table. To prevent the PCI interrupts from being set up
440 * incorrectly, we try to use the ELCR. The sanity check to see if
441 * there is good ELCR data is very simple - IRQ0, 1, 2 and 13 can
442 * never be level sensitive, so we simply see if the ELCR agrees.
443 * If it does, we assume it's valid.
445 if (mpc_default_type == 5) {
446 printk(KERN_INFO "ISA/PCI bus type with no IRQ information... "
447 "falling back to ELCR\n");
449 if (ELCR_trigger(0) || ELCR_trigger(1) || ELCR_trigger(2) ||
451 printk(KERN_ERR "ELCR contains invalid data... "
455 "Using ELCR to identify PCI interrupts\n");
460 for (i = 0; i < 16; i++) {
461 switch (mpc_default_type) {
463 if (i == 0 || i == 13)
464 continue; /* IRQ0 & IRQ13 not connected */
468 continue; /* IRQ2 is never connected */
473 * If the ELCR indicates a level-sensitive interrupt, we
474 * copy that information over to the MP table in the
475 * irqflag field (level sensitive, active high polarity).
478 intsrc.mpc_irqflag = 13;
480 intsrc.mpc_irqflag = 0;
483 intsrc.mpc_srcbusirq = i;
484 intsrc.mpc_dstirq = i ? i : 2; /* IRQ0 to INTIN2 */
485 MP_intsrc_info(&intsrc);
488 intsrc.mpc_irqtype = mp_ExtINT;
489 intsrc.mpc_srcbusirq = 0;
490 intsrc.mpc_dstirq = 0; /* 8259A to INTIN0 */
491 MP_intsrc_info(&intsrc);
496 static inline void __init construct_default_ISA_mptable(int mpc_default_type)
498 struct mpc_config_processor processor;
499 struct mpc_config_bus bus;
500 #ifdef CONFIG_X86_IO_APIC
501 struct mpc_config_ioapic ioapic;
503 struct mpc_config_lintsrc lintsrc;
504 int linttypes[2] = { mp_ExtINT, mp_NMI };
508 * local APIC has default address
510 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
513 * 2 CPUs, numbered 0 & 1.
515 processor.mpc_type = MP_PROCESSOR;
516 /* Either an integrated APIC or a discrete 82489DX. */
517 processor.mpc_apicver = mpc_default_type > 4 ? 0x10 : 0x01;
518 processor.mpc_cpuflag = CPU_ENABLED;
519 processor.mpc_cpufeature = (boot_cpu_data.x86 << 8) |
520 (boot_cpu_data.x86_model << 4) | boot_cpu_data.x86_mask;
521 processor.mpc_featureflag = boot_cpu_data.x86_capability[0];
522 processor.mpc_reserved[0] = 0;
523 processor.mpc_reserved[1] = 0;
524 for (i = 0; i < 2; i++) {
525 processor.mpc_apicid = i;
526 MP_processor_info(&processor);
529 bus.mpc_type = MP_BUS;
531 switch (mpc_default_type) {
533 printk(KERN_ERR "???\nUnknown standard configuration %d\n",
538 memcpy(bus.mpc_bustype, "ISA ", 6);
543 memcpy(bus.mpc_bustype, "EISA ", 6);
547 memcpy(bus.mpc_bustype, "MCA ", 6);
550 if (mpc_default_type > 4) {
552 memcpy(bus.mpc_bustype, "PCI ", 6);
556 #ifdef CONFIG_X86_IO_APIC
557 ioapic.mpc_type = MP_IOAPIC;
558 ioapic.mpc_apicid = 2;
559 ioapic.mpc_apicver = mpc_default_type > 4 ? 0x10 : 0x01;
560 ioapic.mpc_flags = MPC_APIC_USABLE;
561 ioapic.mpc_apicaddr = 0xFEC00000;
562 MP_ioapic_info(&ioapic);
565 * We set up most of the low 16 IO-APIC pins according to MPS rules.
567 construct_default_ioirq_mptable(mpc_default_type);
569 lintsrc.mpc_type = MP_LINTSRC;
570 lintsrc.mpc_irqflag = 0; /* conforming */
571 lintsrc.mpc_srcbusid = 0;
572 lintsrc.mpc_srcbusirq = 0;
573 lintsrc.mpc_destapic = MP_APIC_ALL;
574 for (i = 0; i < 2; i++) {
575 lintsrc.mpc_irqtype = linttypes[i];
576 lintsrc.mpc_destapiclint = i;
577 MP_lintsrc_info(&lintsrc);
581 static struct intel_mp_floating *mpf_found;
584 * Scan the memory blocks for an SMP configuration block.
586 static void __init __get_smp_config(unsigned early)
588 struct intel_mp_floating *mpf = mpf_found;
590 if (acpi_lapic && early)
593 * ACPI supports both logical (e.g. Hyper-Threading) and physical
594 * processors, where MPS only supports physical.
596 if (acpi_lapic && acpi_ioapic) {
597 printk(KERN_INFO "Using ACPI (MADT) for SMP configuration "
600 } else if (acpi_lapic)
601 printk(KERN_INFO "Using ACPI for processor (LAPIC) "
602 "configuration information\n");
604 printk(KERN_INFO "Intel MultiProcessor Specification v1.%d\n",
605 mpf->mpf_specification);
607 if (mpf->mpf_feature2 & (1 << 7)) {
608 printk(KERN_INFO " IMCR and PIC compatibility mode.\n");
611 printk(KERN_INFO " Virtual Wire compatibility mode.\n");
616 * Now see if we need to read further.
618 if (mpf->mpf_feature1 != 0) {
621 * local APIC has default address
623 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
627 printk(KERN_INFO "Default MP configuration #%d\n",
629 construct_default_ISA_mptable(mpf->mpf_feature1);
631 } else if (mpf->mpf_physptr) {
634 * Read the physical hardware table. Anything here will
635 * override the defaults.
637 if (!smp_read_mpc(phys_to_virt(mpf->mpf_physptr), early)) {
638 smp_found_config = 0;
640 "BIOS bug, MP table errors detected!...\n");
641 printk(KERN_ERR "... disabling SMP support. "
642 "(tell your hw vendor)\n");
648 #ifdef CONFIG_X86_IO_APIC
650 * If there are no explicit MP IRQ entries, then we are
651 * broken. We set up most of the low 16 IO-APIC pins to
652 * ISA defaults and hope it will work.
654 if (!mp_irq_entries) {
655 struct mpc_config_bus bus;
657 printk(KERN_ERR "BIOS bug, no explicit IRQ entries, "
658 "using default mptable. "
659 "(tell your hw vendor)\n");
661 bus.mpc_type = MP_BUS;
663 memcpy(bus.mpc_bustype, "ISA ", 6);
666 construct_default_ioirq_mptable(0);
673 printk(KERN_INFO "Processors: %d\n", num_processors);
675 * Only use the first configuration found.
679 void __init early_get_smp_config(void)
684 void __init get_smp_config(void)
689 static int __init smp_scan_config(unsigned long base, unsigned long length,
692 unsigned int *bp = phys_to_virt(base);
693 struct intel_mp_floating *mpf;
695 printk(KERN_DEBUG "Scan SMP from %p for %ld bytes.\n", bp, length);
696 BUILD_BUG_ON(sizeof(*mpf) != 16);
699 mpf = (struct intel_mp_floating *)bp;
700 if ((*bp == SMP_MAGIC_IDENT) &&
701 (mpf->mpf_length == 1) &&
702 !mpf_checksum((unsigned char *)bp, 16) &&
703 ((mpf->mpf_specification == 1)
704 || (mpf->mpf_specification == 4))) {
706 smp_found_config = 1;
709 printk(KERN_INFO "found SMP MP-table at [%p] %08lx\n",
710 mpf, virt_to_phys(mpf));
711 reserve_bootmem(virt_to_phys(mpf), PAGE_SIZE,
713 if (mpf->mpf_physptr) {
715 * We cannot access to MPC table to compute
716 * table size yet, as only few megabytes from
717 * the bottom is mapped now.
718 * PC-9800's MPC table places on the very last
719 * of physical memory; so that simply reserving
720 * PAGE_SIZE from mpg->mpf_physptr yields BUG()
721 * in reserve_bootmem.
723 unsigned long size = PAGE_SIZE;
724 unsigned long end = max_low_pfn * PAGE_SIZE;
725 if (mpf->mpf_physptr + size > end)
726 size = end - mpf->mpf_physptr;
727 reserve_bootmem(mpf->mpf_physptr, size,
735 reserve_bootmem_generic(virt_to_phys(mpf), PAGE_SIZE);
736 if (mpf->mpf_physptr)
737 reserve_bootmem_generic(mpf->mpf_physptr,
748 static void __init __find_smp_config(unsigned reserve)
750 unsigned int address;
753 * FIXME: Linux assumes you have 640K of base ram..
754 * this continues the error...
756 * 1) Scan the bottom 1K for a signature
757 * 2) Scan the top 1K of base RAM
758 * 3) Scan the 64K of bios
760 if (smp_scan_config(0x0, 0x400, reserve) ||
761 smp_scan_config(639 * 0x400, 0x400, reserve) ||
762 smp_scan_config(0xF0000, 0x10000, reserve))
765 * If it is an SMP machine we should know now, unless the
766 * configuration is in an EISA/MCA bus machine with an
767 * extended bios data area.
769 * there is a real-mode segmented pointer pointing to the
770 * 4K EBDA area at 0x40E, calculate and scan it here.
772 * NOTE! There are Linux loaders that will corrupt the EBDA
773 * area, and as such this kind of SMP config may be less
774 * trustworthy, simply because the SMP table may have been
775 * stomped on during early boot. These loaders are buggy and
778 * MP1.4 SPEC states to only scan first 1K of 4K EBDA.
781 address = get_bios_ebda();
783 smp_scan_config(address, 0x400, reserve);
786 void __init early_find_smp_config(void)
788 __find_smp_config(0);
791 void __init find_smp_config(void)
793 __find_smp_config(1);