2 * x86 SMP booting functions
4 * (c) 1995 Alan Cox, Building #3 <alan@redhat.com>
5 * (c) 1998, 1999, 2000 Ingo Molnar <mingo@redhat.com>
7 * Much of the core SMP work is based on previous work by Thomas Radke, to
8 * whom a great many thanks are extended.
10 * Thanks to Intel for making available several different Pentium,
11 * Pentium Pro and Pentium-II/Xeon MP machines.
12 * Original development of Linux SMP code supported by Caldera.
14 * This code is released under the GNU General Public License version 2 or
18 * Felix Koop : NR_CPUS used properly
19 * Jose Renau : Handle single CPU case.
20 * Alan Cox : By repeated request 8) - Total BogoMIPS report.
21 * Greg Wright : Fix for kernel stacks panic.
22 * Erich Boleyn : MP v1.4 and additional changes.
23 * Matthias Sattler : Changes for 2.1 kernel map.
24 * Michel Lespinasse : Changes for 2.1 kernel map.
25 * Michael Chastain : Change trampoline.S to gnu as.
26 * Alan Cox : Dumb bug: 'B' step PPro's are fine
27 * Ingo Molnar : Added APIC timers, based on code
29 * Ingo Molnar : various cleanups and rewrites
30 * Tigran Aivazian : fixed "0.00 in /proc/uptime on SMP" bug.
31 * Maciej W. Rozycki : Bits for genuine 82489DX APICs
32 * Martin J. Bligh : Added support for multi-quad systems
33 * Dave Jones : Report invalid combinations of Athlon CPUs.
34 * Rusty Russell : Hacked into shape for new "hotplug" boot process. */
36 #include <linux/module.h>
37 #include <linux/config.h>
38 #include <linux/init.h>
39 #include <linux/kernel.h>
42 #include <linux/sched.h>
43 #include <linux/kernel_stat.h>
44 #include <linux/smp_lock.h>
45 #include <linux/irq.h>
46 #include <linux/bootmem.h>
48 #include <linux/delay.h>
49 #include <linux/mc146818rtc.h>
50 #include <asm/tlbflush.h>
52 #include <asm/arch_hooks.h>
54 #include <mach_apic.h>
55 #include <mach_wakecpu.h>
56 #include <smpboot_hooks.h>
58 /* Set if we find a B stepping CPU */
59 static int __initdata smp_b_stepping
;
61 /* Number of siblings per CPU package */
62 int smp_num_siblings
= 1;
63 int phys_proc_id
[NR_CPUS
]; /* Package ID of each logical CPU */
65 /* bitmap of online cpus */
66 cpumask_t cpu_online_map
;
68 static cpumask_t cpu_callin_map
;
69 cpumask_t cpu_callout_map
;
70 static cpumask_t smp_commenced_mask
;
72 /* Per CPU bogomips and other parameters */
73 struct cpuinfo_x86 cpu_data
[NR_CPUS
] __cacheline_aligned
;
75 u8 x86_cpu_to_apicid
[NR_CPUS
] =
76 { [0 ... NR_CPUS
-1] = 0xff };
77 EXPORT_SYMBOL(x86_cpu_to_apicid
);
79 /* Set when the idlers are all forked */
80 int smp_threads_ready
;
83 * Trampoline 80x86 program as an array.
86 extern unsigned char trampoline_data
[];
87 extern unsigned char trampoline_end
[];
88 static unsigned char *trampoline_base
;
89 static int trampoline_exec
;
92 * Currently trivial. Write the real->protected mode
93 * bootstrap into the page concerned. The caller
94 * has made sure it's suitably aligned.
97 static unsigned long __init
setup_trampoline(void)
99 memcpy(trampoline_base
, trampoline_data
, trampoline_end
- trampoline_data
);
100 return virt_to_phys(trampoline_base
);
104 * We are called very early to get the low memory for the
105 * SMP bootup trampoline page.
107 void __init
smp_alloc_memory(void)
109 trampoline_base
= (void *) alloc_bootmem_low_pages(PAGE_SIZE
);
111 * Has to be in very low memory so we can execute
114 if (__pa(trampoline_base
) >= 0x9F000)
117 * Make the SMP trampoline executable:
119 trampoline_exec
= set_kernel_exec((unsigned long)trampoline_base
, 1);
123 * The bootstrap kernel entry code has set these up. Save them for
127 static void __init
smp_store_cpu_info(int id
)
129 struct cpuinfo_x86
*c
= cpu_data
+ id
;
135 * Mask B, Pentium, but not Pentium MMX
137 if (c
->x86_vendor
== X86_VENDOR_INTEL
&&
139 c
->x86_mask
>= 1 && c
->x86_mask
<= 4 &&
142 * Remember we have B step Pentia with bugs
147 * Certain Athlons might work (for various values of 'work') in SMP
148 * but they are not certified as MP capable.
150 if ((c
->x86_vendor
== X86_VENDOR_AMD
) && (c
->x86
== 6)) {
152 /* Athlon 660/661 is valid. */
153 if ((c
->x86_model
==6) && ((c
->x86_mask
==0) || (c
->x86_mask
==1)))
156 /* Duron 670 is valid */
157 if ((c
->x86_model
==7) && (c
->x86_mask
==0))
161 * Athlon 662, Duron 671, and Athlon >model 7 have capability bit.
162 * It's worth noting that the A5 stepping (662) of some Athlon XP's
163 * have the MP bit set.
164 * See http://www.heise.de/newsticker/data/jow-18.10.01-000 for more.
166 if (((c
->x86_model
==6) && (c
->x86_mask
>=2)) ||
167 ((c
->x86_model
==7) && (c
->x86_mask
>=1)) ||
172 /* If we get here, it's not a certified SMP capable AMD system. */
173 tainted
|= TAINT_UNSAFE_SMP
;
181 * TSC synchronization.
183 * We first check whether all CPUs have their TSC's synchronized,
184 * then we print a warning if not, and always resync.
187 static atomic_t tsc_start_flag
= ATOMIC_INIT(0);
188 static atomic_t tsc_count_start
= ATOMIC_INIT(0);
189 static atomic_t tsc_count_stop
= ATOMIC_INIT(0);
190 static unsigned long long tsc_values
[NR_CPUS
];
194 static void __init
synchronize_tsc_bp (void)
197 unsigned long long t0
;
198 unsigned long long sum
, avg
;
200 unsigned long one_usec
;
203 printk(KERN_INFO
"checking TSC synchronization across %u CPUs: ", num_booting_cpus());
205 /* convert from kcyc/sec to cyc/usec */
206 one_usec
= cpu_khz
/ 1000;
208 atomic_set(&tsc_start_flag
, 1);
212 * We loop a few times to get a primed instruction cache,
213 * then the last pass is more or less synchronized and
214 * the BP and APs set their cycle counters to zero all at
215 * once. This reduces the chance of having random offsets
216 * between the processors, and guarantees that the maximum
217 * delay between the cycle counters is never bigger than
218 * the latency of information-passing (cachelines) between
221 for (i
= 0; i
< NR_LOOPS
; i
++) {
223 * all APs synchronize but they loop on '== num_cpus'
225 while (atomic_read(&tsc_count_start
) != num_booting_cpus()-1)
227 atomic_set(&tsc_count_stop
, 0);
230 * this lets the APs save their current TSC:
232 atomic_inc(&tsc_count_start
);
234 rdtscll(tsc_values
[smp_processor_id()]);
236 * We clear the TSC in the last loop:
242 * Wait for all APs to leave the synchronization point:
244 while (atomic_read(&tsc_count_stop
) != num_booting_cpus()-1)
246 atomic_set(&tsc_count_start
, 0);
248 atomic_inc(&tsc_count_stop
);
252 for (i
= 0; i
< NR_CPUS
; i
++) {
253 if (cpu_isset(i
, cpu_callout_map
)) {
259 do_div(avg
, num_booting_cpus());
262 for (i
= 0; i
< NR_CPUS
; i
++) {
263 if (!cpu_isset(i
, cpu_callout_map
))
265 delta
= tsc_values
[i
] - avg
;
269 * We report bigger than 2 microseconds clock differences.
271 if (delta
> 2*one_usec
) {
278 do_div(realdelta
, one_usec
);
279 if (tsc_values
[i
] < avg
)
280 realdelta
= -realdelta
;
282 printk(KERN_INFO
"CPU#%d had %ld usecs TSC skew, fixed it up.\n", i
, realdelta
);
291 static void __init
synchronize_tsc_ap (void)
296 * Not every cpu is online at the time
297 * this gets called, so we first wait for the BP to
298 * finish SMP initialization:
300 while (!atomic_read(&tsc_start_flag
)) mb();
302 for (i
= 0; i
< NR_LOOPS
; i
++) {
303 atomic_inc(&tsc_count_start
);
304 while (atomic_read(&tsc_count_start
) != num_booting_cpus())
307 rdtscll(tsc_values
[smp_processor_id()]);
311 atomic_inc(&tsc_count_stop
);
312 while (atomic_read(&tsc_count_stop
) != num_booting_cpus()) mb();
317 extern void calibrate_delay(void);
319 static atomic_t init_deasserted
;
321 void __init
smp_callin(void)
324 unsigned long timeout
;
327 * If waken up by an INIT in an 82489DX configuration
328 * we may get here before an INIT-deassert IPI reaches
329 * our local APIC. We have to wait for the IPI or we'll
330 * lock up on an APIC access.
332 wait_for_init_deassert(&init_deasserted
);
335 * (This works even if the APIC is not enabled.)
337 phys_id
= GET_APIC_ID(apic_read(APIC_ID
));
338 cpuid
= smp_processor_id();
339 if (cpu_isset(cpuid
, cpu_callin_map
)) {
340 printk("huh, phys CPU#%d, CPU#%d already present??\n",
344 Dprintk("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid
, phys_id
);
347 * STARTUP IPIs are fragile beasts as they might sometimes
348 * trigger some glue motherboard logic. Complete APIC bus
349 * silence for 1 second, this overestimates the time the
350 * boot CPU is spending to send the up to 2 STARTUP IPIs
351 * by a factor of two. This should be enough.
355 * Waiting 2s total for startup (udelay is not yet working)
357 timeout
= jiffies
+ 2*HZ
;
358 while (time_before(jiffies
, timeout
)) {
360 * Has the boot CPU finished it's STARTUP sequence?
362 if (cpu_isset(cpuid
, cpu_callout_map
))
367 if (!time_before(jiffies
, timeout
)) {
368 printk("BUG: CPU%d started up but did not get a callout!\n",
374 * the boot CPU has finished the init stage and is spinning
375 * on callin_map until we finish. We are free to set up this
376 * CPU, first the APIC. (this is probably redundant on most
380 Dprintk("CALLIN, before setup_local_APIC().\n");
381 smp_callin_clear_local_apic();
383 map_cpu_to_logical_apicid();
391 Dprintk("Stack at about %p\n",&cpuid
);
394 * Save our processor parameters
396 smp_store_cpu_info(cpuid
);
398 disable_APIC_timer();
401 * Allow the master to continue.
403 cpu_set(cpuid
, cpu_callin_map
);
406 * Synchronize the TSC with the BP
408 if (cpu_has_tsc
&& cpu_khz
)
409 synchronize_tsc_ap();
414 extern int cpu_idle(void);
417 * Activate a secondary processor.
419 int __init
start_secondary(void *unused
)
422 * Dont put anything before smp_callin(), SMP
423 * booting is too fragile that we want to limit the
424 * things done here to the most necessary things.
428 while (!cpu_isset(smp_processor_id(), smp_commenced_mask
))
430 setup_secondary_APIC_clock();
431 if (nmi_watchdog
== NMI_IO_APIC
) {
432 disable_8259A_irq(0);
433 enable_NMI_through_LVT0(NULL
);
438 * low-memory mappings have been cleared, flush them from
439 * the local TLBs too.
442 cpu_set(smp_processor_id(), cpu_online_map
);
448 * Everything has been set up for the secondary
449 * CPUs - they just need to reload everything
450 * from the task structure
451 * This function must not return.
453 void __init
initialize_secondary(void)
456 * We don't actually need to load the full TSS,
457 * basically just the stack pointer and the eip.
464 :"r" (current
->thread
.esp
),"r" (current
->thread
.eip
));
474 /* which logical CPUs are on which nodes */
475 cpumask_t node_2_cpu_mask
[MAX_NUMNODES
] =
476 { [0 ... MAX_NUMNODES
-1] = CPU_MASK_NONE
};
477 /* which node each logical CPU is on */
478 int cpu_2_node
[NR_CPUS
] = { [0 ... NR_CPUS
-1] = 0 };
479 EXPORT_SYMBOL(cpu_2_node
);
481 /* set up a mapping between cpu and node. */
482 static inline void map_cpu_to_node(int cpu
, int node
)
484 printk("Mapping cpu %d to node %d\n", cpu
, node
);
485 cpu_set(cpu
, node_2_cpu_mask
[node
]);
486 cpu_2_node
[cpu
] = node
;
489 /* undo a mapping between cpu and node. */
490 static inline void unmap_cpu_to_node(int cpu
)
494 printk("Unmapping cpu %d from all nodes\n", cpu
);
495 for (node
= 0; node
< MAX_NUMNODES
; node
++)
496 cpu_clear(cpu
, node_2_cpu_mask
[node
]);
499 #else /* !CONFIG_NUMA */
501 #define map_cpu_to_node(cpu, node) ({})
502 #define unmap_cpu_to_node(cpu) ({})
504 #endif /* CONFIG_NUMA */
506 u8 cpu_2_logical_apicid
[NR_CPUS
] = { [0 ... NR_CPUS
-1] = BAD_APICID
};
508 void map_cpu_to_logical_apicid(void)
510 int cpu
= smp_processor_id();
511 int apicid
= logical_smp_processor_id();
513 cpu_2_logical_apicid
[cpu
] = apicid
;
514 map_cpu_to_node(cpu
, apicid_to_node(apicid
));
517 void unmap_cpu_to_logical_apicid(int cpu
)
519 cpu_2_logical_apicid
[cpu
] = BAD_APICID
;
520 unmap_cpu_to_node(cpu
);
524 static inline void __inquire_remote_apic(int apicid
)
526 int i
, regs
[] = { APIC_ID
>> 4, APIC_LVR
>> 4, APIC_SPIV
>> 4 };
527 char *names
[] = { "ID", "VERSION", "SPIV" };
530 printk("Inquiring remote APIC #%d...\n", apicid
);
532 for (i
= 0; i
< sizeof(regs
) / sizeof(*regs
); i
++) {
533 printk("... APIC #%d %s: ", apicid
, names
[i
]);
538 apic_wait_icr_idle();
540 apic_write_around(APIC_ICR2
, SET_APIC_DEST_FIELD(apicid
));
541 apic_write_around(APIC_ICR
, APIC_DM_REMRD
| regs
[i
]);
546 status
= apic_read(APIC_ICR
) & APIC_ICR_RR_MASK
;
547 } while (status
== APIC_ICR_RR_INPROG
&& timeout
++ < 1000);
550 case APIC_ICR_RR_VALID
:
551 status
= apic_read(APIC_RRR
);
552 printk("%08x\n", status
);
561 #ifdef WAKE_SECONDARY_VIA_NMI
563 * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
564 * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
565 * won't ... remember to clear down the APIC, etc later.
568 wakeup_secondary_cpu(int logical_apicid
, unsigned long start_eip
)
570 unsigned long send_status
= 0, accept_status
= 0;
574 apic_write_around(APIC_ICR2
, SET_APIC_DEST_FIELD(logical_apicid
));
576 /* Boot on the stack */
577 /* Kick the second */
578 apic_write_around(APIC_ICR
, APIC_DM_NMI
| APIC_DEST_LOGICAL
);
580 Dprintk("Waiting for send to finish...\n");
585 send_status
= apic_read(APIC_ICR
) & APIC_ICR_BUSY
;
586 } while (send_status
&& (timeout
++ < 1000));
589 * Give the other CPU some time to accept the IPI.
593 * Due to the Pentium erratum 3AP.
595 maxlvt
= get_maxlvt();
597 apic_read_around(APIC_SPIV
);
598 apic_write(APIC_ESR
, 0);
600 accept_status
= (apic_read(APIC_ESR
) & 0xEF);
601 Dprintk("NMI sent.\n");
604 printk("APIC never delivered???\n");
606 printk("APIC delivery error (%lx).\n", accept_status
);
608 return (send_status
| accept_status
);
610 #endif /* WAKE_SECONDARY_VIA_NMI */
612 #ifdef WAKE_SECONDARY_VIA_INIT
614 wakeup_secondary_cpu(int phys_apicid
, unsigned long start_eip
)
616 unsigned long send_status
= 0, accept_status
= 0;
617 int maxlvt
, timeout
, num_starts
, j
;
620 * Be paranoid about clearing APIC errors.
622 if (APIC_INTEGRATED(apic_version
[phys_apicid
])) {
623 apic_read_around(APIC_SPIV
);
624 apic_write(APIC_ESR
, 0);
628 Dprintk("Asserting INIT.\n");
631 * Turn INIT on target chip
633 apic_write_around(APIC_ICR2
, SET_APIC_DEST_FIELD(phys_apicid
));
638 apic_write_around(APIC_ICR
, APIC_INT_LEVELTRIG
| APIC_INT_ASSERT
641 Dprintk("Waiting for send to finish...\n");
646 send_status
= apic_read(APIC_ICR
) & APIC_ICR_BUSY
;
647 } while (send_status
&& (timeout
++ < 1000));
651 Dprintk("Deasserting INIT.\n");
654 apic_write_around(APIC_ICR2
, SET_APIC_DEST_FIELD(phys_apicid
));
657 apic_write_around(APIC_ICR
, APIC_INT_LEVELTRIG
| APIC_DM_INIT
);
659 Dprintk("Waiting for send to finish...\n");
664 send_status
= apic_read(APIC_ICR
) & APIC_ICR_BUSY
;
665 } while (send_status
&& (timeout
++ < 1000));
667 atomic_set(&init_deasserted
, 1);
670 * Should we send STARTUP IPIs ?
672 * Determine this based on the APIC version.
673 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
675 if (APIC_INTEGRATED(apic_version
[phys_apicid
]))
681 * Run STARTUP IPI loop.
683 Dprintk("#startup loops: %d.\n", num_starts
);
685 maxlvt
= get_maxlvt();
687 for (j
= 1; j
<= num_starts
; j
++) {
688 Dprintk("Sending STARTUP #%d.\n",j
);
689 apic_read_around(APIC_SPIV
);
690 apic_write(APIC_ESR
, 0);
692 Dprintk("After apic_write.\n");
699 apic_write_around(APIC_ICR2
, SET_APIC_DEST_FIELD(phys_apicid
));
701 /* Boot on the stack */
702 /* Kick the second */
703 apic_write_around(APIC_ICR
, APIC_DM_STARTUP
704 | (start_eip
>> 12));
707 * Give the other CPU some time to accept the IPI.
711 Dprintk("Startup point 1.\n");
713 Dprintk("Waiting for send to finish...\n");
718 send_status
= apic_read(APIC_ICR
) & APIC_ICR_BUSY
;
719 } while (send_status
&& (timeout
++ < 1000));
722 * Give the other CPU some time to accept the IPI.
726 * Due to the Pentium erratum 3AP.
729 apic_read_around(APIC_SPIV
);
730 apic_write(APIC_ESR
, 0);
732 accept_status
= (apic_read(APIC_ESR
) & 0xEF);
733 if (send_status
|| accept_status
)
736 Dprintk("After Startup.\n");
739 printk("APIC never delivered???\n");
741 printk("APIC delivery error (%lx).\n", accept_status
);
743 return (send_status
| accept_status
);
745 #endif /* WAKE_SECONDARY_VIA_INIT */
747 extern cpumask_t cpu_initialized
;
749 static int __init
do_boot_cpu(int apicid
)
751 * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
752 * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
753 * Returns zero if CPU booted OK, else error code from wakeup_secondary_cpu.
756 struct task_struct
*idle
;
757 unsigned long boot_error
;
759 unsigned long start_eip
;
760 unsigned short nmi_high
= 0, nmi_low
= 0;
764 * We can't use kernel_thread since we must avoid to
765 * reschedule the child.
767 idle
= fork_idle(cpu
);
769 panic("failed fork for CPU %d", cpu
);
770 idle
->thread
.eip
= (unsigned long) start_secondary
;
771 /* start_eip had better be page-aligned! */
772 start_eip
= setup_trampoline();
774 /* So we see what's up */
775 printk("Booting processor %d/%d eip %lx\n", cpu
, apicid
, start_eip
);
776 /* Stack for startup_32 can be just as for start_secondary onwards */
777 stack_start
.esp
= (void *) idle
->thread
.esp
;
782 * This grunge runs the startup process for
783 * the targeted processor.
786 atomic_set(&init_deasserted
, 0);
788 Dprintk("Setting warm reset code and vector.\n");
790 store_NMI_vector(&nmi_high
, &nmi_low
);
792 smpboot_setup_warm_reset_vector(start_eip
);
795 * Starting actual IPI sequence...
797 boot_error
= wakeup_secondary_cpu(apicid
, start_eip
);
801 * allow APs to start initializing.
803 Dprintk("Before Callout %d.\n", cpu
);
804 cpu_set(cpu
, cpu_callout_map
);
805 Dprintk("After Callout %d.\n", cpu
);
808 * Wait 5s total for a response
810 for (timeout
= 0; timeout
< 50000; timeout
++) {
811 if (cpu_isset(cpu
, cpu_callin_map
))
812 break; /* It has booted */
816 if (cpu_isset(cpu
, cpu_callin_map
)) {
817 /* number CPUs logically, starting from 1 (BSP is 0) */
819 printk("CPU%d: ", cpu
);
820 print_cpu_info(&cpu_data
[cpu
]);
821 Dprintk("CPU has booted.\n");
824 if (*((volatile unsigned char *)trampoline_base
)
826 /* trampoline started but...? */
827 printk("Stuck ??\n");
829 /* trampoline code not run */
830 printk("Not responding.\n");
831 inquire_remote_apic(apicid
);
834 x86_cpu_to_apicid
[cpu
] = apicid
;
836 /* Try to put things back the way they were before ... */
837 unmap_cpu_to_logical_apicid(cpu
);
838 cpu_clear(cpu
, cpu_callout_map
); /* was set here (do_boot_cpu()) */
839 cpu_clear(cpu
, cpu_initialized
); /* was set by cpu_init() */
843 /* mark "stuck" area as not stuck */
844 *((volatile unsigned long *)trampoline_base
) = 0;
849 cycles_t cacheflush_time
;
850 unsigned long cache_decay_ticks
;
852 static void smp_tune_scheduling (void)
854 unsigned long cachesize
; /* kB */
855 unsigned long bandwidth
= 350; /* MB/s */
857 * Rough estimation for SMP scheduling, this is the number of
858 * cycles it takes for a fully memory-limited process to flush
859 * the SMP-local cache.
861 * (For a P5 this pretty much means we will choose another idle
862 * CPU almost always at wakeup time (this is due to the small
863 * L1 cache), on PIIs it's around 50-100 usecs, depending on
869 * this basically disables processor-affinity
870 * scheduling on SMP without a TSC.
875 cachesize
= boot_cpu_data
.x86_cache_size
;
876 if (cachesize
== -1) {
877 cachesize
= 16; /* Pentiums, 2x8kB cache */
881 cacheflush_time
= (cpu_khz
>>10) * (cachesize
<<10) / bandwidth
;
884 cache_decay_ticks
= (long)cacheflush_time
/cpu_khz
+ 1;
886 printk("per-CPU timeslice cutoff: %ld.%02ld usecs.\n",
887 (long)cacheflush_time
/(cpu_khz
/1000),
888 ((long)cacheflush_time
*100/(cpu_khz
/1000)) % 100);
889 printk("task migration cache decay timeout: %ld msecs.\n",
894 * Cycle through the processors sending APIC IPIs to boot each.
897 static int boot_cpu_logical_apicid
;
898 /* Where the IO area was mapped on multiquad, always 0 otherwise */
901 cpumask_t cpu_sibling_map
[NR_CPUS
] __cacheline_aligned
;
903 static void __init
smp_boot_cpus(unsigned int max_cpus
)
905 int apicid
, cpu
, bit
, kicked
;
906 unsigned long bogosum
= 0;
909 * Setup boot CPU information
911 smp_store_cpu_info(0); /* Final full version of the data */
912 printk("CPU%d: ", 0);
913 print_cpu_info(&cpu_data
[0]);
915 boot_cpu_physical_apicid
= GET_APIC_ID(apic_read(APIC_ID
));
916 boot_cpu_logical_apicid
= logical_smp_processor_id();
917 x86_cpu_to_apicid
[0] = boot_cpu_physical_apicid
;
919 current_thread_info()->cpu
= 0;
920 smp_tune_scheduling();
921 cpus_clear(cpu_sibling_map
[0]);
922 cpu_set(0, cpu_sibling_map
[0]);
925 * If we couldn't find an SMP configuration at boot time,
926 * get out of here now!
928 if (!smp_found_config
&& !acpi_lapic
) {
929 printk(KERN_NOTICE
"SMP motherboard not detected.\n");
930 smpboot_clear_io_apic_irqs();
931 phys_cpu_present_map
= physid_mask_of_physid(0);
932 if (APIC_init_uniprocessor())
933 printk(KERN_NOTICE
"Local APIC not detected."
934 " Using dummy APIC emulation.\n");
935 map_cpu_to_logical_apicid();
940 * Should not be necessary because the MP table should list the boot
941 * CPU too, but we do it for the sake of robustness anyway.
942 * Makes no sense to do this check in clustered apic mode, so skip it
944 if (!check_phys_apicid_present(boot_cpu_physical_apicid
)) {
945 printk("weird, boot CPU (#%d) not listed by the BIOS.\n",
946 boot_cpu_physical_apicid
);
947 physid_set(hard_smp_processor_id(), phys_cpu_present_map
);
951 * If we couldn't find a local APIC, then get out of here now!
953 if (APIC_INTEGRATED(apic_version
[boot_cpu_physical_apicid
]) && !cpu_has_apic
) {
954 printk(KERN_ERR
"BIOS bug, local APIC #%d not detected!...\n",
955 boot_cpu_physical_apicid
);
956 printk(KERN_ERR
"... forcing use of dummy APIC emulation. (tell your hw vendor)\n");
957 smpboot_clear_io_apic_irqs();
958 phys_cpu_present_map
= physid_mask_of_physid(0);
965 * If SMP should be disabled, then really disable it!
968 smp_found_config
= 0;
969 printk(KERN_INFO
"SMP mode deactivated, forcing use of dummy APIC emulation.\n");
970 smpboot_clear_io_apic_irqs();
971 phys_cpu_present_map
= physid_mask_of_physid(0);
977 map_cpu_to_logical_apicid();
980 setup_portio_remap();
983 * Scan the CPU present map and fire up the other CPUs via do_boot_cpu
985 * In clustered apic mode, phys_cpu_present_map is a constructed thus:
986 * bits 0-3 are quad0, 4-7 are quad1, etc. A perverse twist on the
989 Dprintk("CPU present map: %lx\n", physids_coerce(phys_cpu_present_map
));
992 for (bit
= 0; kicked
< NR_CPUS
&& bit
< MAX_APICS
; bit
++) {
993 apicid
= cpu_present_to_apicid(bit
);
995 * Don't even attempt to start the boot CPU!
997 if ((apicid
== boot_cpu_apicid
) || (apicid
== BAD_APICID
))
1000 if (!check_apicid_present(bit
))
1002 if (max_cpus
<= cpucount
+1)
1005 if (do_boot_cpu(apicid
))
1006 printk("CPU #%d not responding - cannot use it.\n",
1013 * Cleanup possible dangling ends...
1015 smpboot_restore_warm_reset_vector();
1018 * Allow the user to impress friends.
1020 Dprintk("Before bogomips.\n");
1021 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++)
1022 if (cpu_isset(cpu
, cpu_callout_map
))
1023 bogosum
+= cpu_data
[cpu
].loops_per_jiffy
;
1025 "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
1027 bogosum
/(500000/HZ
),
1028 (bogosum
/(5000/HZ
))%100);
1030 Dprintk("Before bogocount - setting activated=1.\n");
1033 printk(KERN_WARNING
"WARNING: SMP operation may be unreliable with B stepping processors.\n");
1036 * Don't taint if we are running SMP kernel on a single non-MP
1039 if (tainted
& TAINT_UNSAFE_SMP
) {
1041 printk (KERN_INFO
"WARNING: This combination of AMD processors is not suitable for SMP.\n");
1043 tainted
&= ~TAINT_UNSAFE_SMP
;
1046 Dprintk("Boot done.\n");
1049 * construct cpu_sibling_map[], so that we can tell sibling CPUs
1052 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++)
1053 cpus_clear(cpu_sibling_map
[cpu
]);
1055 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
1058 if (!cpu_isset(cpu
, cpu_callout_map
))
1061 if (smp_num_siblings
> 1) {
1062 for (i
= 0; i
< NR_CPUS
; i
++) {
1063 if (!cpu_isset(i
, cpu_callout_map
))
1065 if (phys_proc_id
[cpu
] == phys_proc_id
[i
]) {
1067 cpu_set(i
, cpu_sibling_map
[cpu
]);
1072 cpu_set(cpu
, cpu_sibling_map
[cpu
]);
1075 if (siblings
!= smp_num_siblings
)
1076 printk(KERN_WARNING
"WARNING: %d siblings found for CPU%d, should be %d\n", siblings
, cpu
, smp_num_siblings
);
1079 if (nmi_watchdog
== NMI_LOCAL_APIC
)
1080 check_nmi_watchdog();
1082 smpboot_setup_io_apic();
1084 setup_boot_APIC_clock();
1087 * Synchronize the TSC with the AP
1089 if (cpu_has_tsc
&& cpucount
&& cpu_khz
)
1090 synchronize_tsc_bp();
1093 /* These are wrappers to interface to the new boot process. Someone
1094 who understands all this stuff should rewrite it properly. --RR 15/Jul/02 */
1095 void __init
smp_prepare_cpus(unsigned int max_cpus
)
1097 smp_boot_cpus(max_cpus
);
1100 void __devinit
smp_prepare_boot_cpu(void)
1102 cpu_set(smp_processor_id(), cpu_online_map
);
1103 cpu_set(smp_processor_id(), cpu_callout_map
);
1106 int __devinit
__cpu_up(unsigned int cpu
)
1108 /* This only works at boot for x86. See "rewrite" above. */
1109 if (cpu_isset(cpu
, smp_commenced_mask
)) {
1114 /* In case one didn't come up */
1115 if (!cpu_isset(cpu
, cpu_callin_map
)) {
1121 /* Unleash the CPU! */
1122 cpu_set(cpu
, smp_commenced_mask
);
1123 while (!cpu_isset(cpu
, cpu_online_map
))
1128 void __init
smp_cpus_done(unsigned int max_cpus
)
1130 #ifdef CONFIG_X86_IO_APIC
1131 setup_ioapic_dest();
1135 * Disable executability of the SMP trampoline:
1137 set_kernel_exec((unsigned long)trampoline_base
, trampoline_exec
);
1140 void __init
smp_intr_init(void)
1143 * IRQ0 must be given a fixed assignment and initialized,
1144 * because it's used before the IO-APIC is set up.
1146 set_intr_gate(FIRST_DEVICE_VECTOR
, interrupt
[0]);
1149 * The reschedule interrupt is a CPU-to-CPU reschedule-helper
1150 * IPI, driven by wakeup.
1152 set_intr_gate(RESCHEDULE_VECTOR
, reschedule_interrupt
);
1154 /* IPI for invalidation */
1155 set_intr_gate(INVALIDATE_TLB_VECTOR
, invalidate_interrupt
);
1157 /* IPI for generic function call */
1158 set_intr_gate(CALL_FUNCTION_VECTOR
, call_function_interrupt
);