/* smp.c: Sparc SMP support.
*
* Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
*/
#include <asm/head.h>
#include <asm/ptrace.h>
#include <linux/kernel.h>
#include <linux/tasks.h>
#include <linux/smp.h>
#include <asm/delay.h>
#include <asm/irq.h>
#include <asm/page.h>
#include <asm/pgtable.h>
#include <asm/oplib.h>
extern ctxd_t *srmmu_ctx_table_phys;
extern int linux_num_cpus;
struct tlog {
unsigned long pc;
unsigned long psr;
};
struct tlog trap_log[4][256];
unsigned long trap_log_ent[4] = { 0, 0, 0, 0, };
extern void calibrate_delay(void);
volatile unsigned long stuck_pc = 0;
volatile int smp_processors_ready = 0;
int smp_found_config = 0;
unsigned long cpu_present_map = 0;
int smp_num_cpus = 1;
int smp_threads_ready=0;
unsigned char mid_xlate[NR_CPUS] = { 0, 0, 0, 0, };
volatile unsigned long cpu_callin_map[NR_CPUS] = {0,};
volatile unsigned long smp_invalidate_needed[NR_CPUS] = { 0, };
volatile unsigned long smp_spinning[NR_CPUS] = { 0, };
struct cpuinfo_sparc cpu_data[NR_CPUS];
unsigned char boot_cpu_id = 0;
static int smp_activated = 0;
static volatile unsigned char smp_cpu_in_msg[NR_CPUS];
static volatile unsigned long smp_msg_data;
static volatile int smp_src_cpu;
static volatile int smp_msg_id;
volatile int cpu_number_map[NR_CPUS];
volatile int cpu_logical_map[NR_CPUS];
/* The only guaranteed locking primitive available on all Sparc
* processors is 'ldstub [%reg + immediate], %dest_reg' which atomically
* places the current byte at the effective address into dest_reg and
* places 0xff there afterwards. Pretty lame locking primitive
* compared to the Alpha and the intel no? Most Sparcs have 'swap'
* instruction which is much better...
*/
klock_t kernel_flag = KLOCK_CLEAR;
volatile unsigned char active_kernel_processor = NO_PROC_ID;
volatile unsigned long kernel_counter = 0;
volatile unsigned long syscall_count = 0;
volatile unsigned long ipi_count;
#ifdef __SMP_PROF__
volatile unsigned long smp_spins[NR_CPUS]={0};
volatile unsigned long smp_spins_syscall[NR_CPUS]={0};
volatile unsigned long smp_spins_syscall_cur[NR_CPUS]={0};
volatile unsigned long smp_spins_sys_idle[NR_CPUS]={0};
volatile unsigned long smp_idle_count[1+NR_CPUS]={0,};
#endif
#if defined (__SMP_PROF__)
volatile unsigned long smp_idle_map=0;
#endif
volatile unsigned long smp_proc_in_lock[NR_CPUS] = {0,};
volatile int smp_process_available=0;
/*#define SMP_DEBUG*/
#ifdef SMP_DEBUG
#define SMP_PRINTK(x) printk x
#else
#define SMP_PRINTK(x)
#endif
static volatile int smp_commenced = 0;
static char smp_buf[512];
char *smp_info(void)
{
sprintf(smp_buf,
"\n CPU0\t\tCPU1\t\tCPU2\t\tCPU3\n"
"State: %s\t\t%s\t\t%s\t\t%s\n"
"Lock: %08lx\t\t%08lx\t%08lx\t%08lx\n"
"\n"
"klock: %x\n",
(cpu_present_map & 1) ? ((active_kernel_processor == 0) ? "akp" : "online") : "offline",
(cpu_present_map & 2) ? ((active_kernel_processor == 1) ? "akp" : "online") : "offline",
(cpu_present_map & 4) ? ((active_kernel_processor == 2) ? "akp" : "online") : "offline",
(cpu_present_map & 8) ? ((active_kernel_processor == 3) ? "akp" : "online") : "offline",
smp_proc_in_lock[0], smp_proc_in_lock[1], smp_proc_in_lock[2],
smp_proc_in_lock[3],
kernel_flag);
return smp_buf;
}
static inline unsigned long swap(volatile unsigned long *ptr, unsigned long val)
{
__asm__ __volatile__("swap [%1], %0\n\t" :
"=&r" (val), "=&r" (ptr) :
"0" (val), "1" (ptr));
return val;
}
/*
* The bootstrap kernel entry code has set these up. Save them for
* a given CPU
*/
void smp_store_cpu_info(int id)
{
cpu_data[id].udelay_val = loops_per_sec; /* this is it on sparc. */
}
/*
* Architecture specific routine called by the kernel just before init is
* fired off. This allows the BP to have everything in order [we hope].
* At the end of this all the AP's will hit the system scheduling and off
* we go. Each AP will load the system gdt's and jump through the kernel
* init into idle(). At this point the scheduler will one day take over
* and give them jobs to do. smp_callin is a standard routine
* we use to track CPU's as they power up.
*/
void smp_commence(void)
{
/*
* Lets the callin's below out of their loop.
*/
local_flush_cache_all();
local_flush_tlb_all();
smp_commenced = 1;
local_flush_cache_all();
local_flush_tlb_all();
}
void smp_callin(void)
{
int cpuid = smp_processor_id();
sti();
local_flush_cache_all();
local_flush_tlb_all();
calibrate_delay();
smp_store_cpu_info(cpuid);
local_flush_cache_all();
local_flush_tlb_all();
cli();
/* Allow master to continue. */
swap((unsigned long *)&cpu_callin_map[cpuid], 1);
local_flush_cache_all();
local_flush_tlb_all();
while(!smp_commenced)
barrier();
local_flush_cache_all();
local_flush_tlb_all();
/* Fix idle thread fields. */
current->mm->mmap->vm_page_prot = PAGE_SHARED;
current->mm->mmap->vm_start = KERNBASE;
current->mm->mmap->vm_end = init_task.mm->mmap->vm_end;
local_flush_cache_all();
local_flush_tlb_all();
sti();
}
void cpu_panic(void)
{
printk("CPU[%d]: Returns from cpu_idle!\n", smp_processor_id());
panic("SMP bolixed\n");
}
/*
* Cycle through the processors asking the PROM to start each one.
*/
extern struct prom_cpuinfo linux_cpus[NCPUS];
static struct linux_prom_registers penguin_ctable;
void smp_boot_cpus(void)
{
int cpucount = 0;
int i = 0;
printk("Entering SparclinuxMultiPenguin(SMP) Mode...\n");
penguin_ctable.which_io = 0;
penguin_ctable.phys_addr = (char *) srmmu_ctx_table_phys;
penguin_ctable.reg_size = 0;
sti();
cpu_present_map |= (1 << smp_processor_id());
cpu_present_map = 0;
for(i=0; i < linux_num_cpus; i++)
cpu_present_map |= (1<<i);
for(i=0; i < NR_CPUS; i++)
cpu_number_map[i] = -1;
for(i=0; i < NR_CPUS; i++)
cpu_logical_map[i] = -1;
mid_xlate[boot_cpu_id] = (linux_cpus[boot_cpu_id].mid & ~8);
cpu_number_map[boot_cpu_id] = 0;
cpu_logical_map[0] = boot_cpu_id;
active_kernel_processor = boot_cpu_id;
smp_store_cpu_info(boot_cpu_id);
set_irq_udt(0);
local_flush_cache_all();
if(linux_num_cpus == 1)
return; /* Not an MP box. */
for(i = 0; i < NR_CPUS; i++) {
if(i == boot_cpu_id)
continue;
if(cpu_present_map & (1 << i)) {
extern unsigned long sparc_cpu_startup;
unsigned long *entry = &sparc_cpu_startup;
int timeout;
/* See trampoline.S for details... */
entry += ((i-1) * 6);
/* whirrr, whirrr, whirrrrrrrrr... */
printk("Starting CPU %d at %p\n", i, entry);
mid_xlate[i] = (linux_cpus[i].mid & ~8);
local_flush_cache_all();
prom_startcpu(linux_cpus[i].prom_node,
&penguin_ctable, 0, (char *)entry);
/* wheee... it's going... */
for(timeout = 0; timeout < 5000000; timeout++) {
if(cpu_callin_map[i])
break;
udelay(100);
}
if(cpu_callin_map[i]) {
/* Another "Red Snapper". */
cpucount++;
cpu_number_map[i] = i;
cpu_logical_map[i] = i;
} else {
printk("Penguin %d is stuck in the bottle.\n", i);
}
}
if(!(cpu_callin_map[i])) {
cpu_present_map &= ~(1 << i);
cpu_number_map[i] = -1;
}
}
local_flush_cache_all();
if(cpucount == 0) {
printk("Error: only one Penguin found.\n");
cpu_present_map = (1 << smp_processor_id());
} else {
unsigned long bogosum = 0;
for(i = 0; i < NR_CPUS; i++) {
if(cpu_present_map & (1 << i))
bogosum += cpu_data[i].udelay_val;
}
printk("Total of %d Penguins activated (%lu.%02lu PenguinMIPS).\n",
cpucount + 1,
(bogosum + 2500)/500000,
((bogosum + 2500)/5000)%100);
smp_activated = 1;
smp_num_cpus = cpucount + 1;
}
smp_processors_ready = 1;
}
static inline void send_ipi(unsigned long target_map, int irq)
{
int i;
for(i = 0; i < 4; i++) {
if((1<<i) & target_map)
set_cpu_int(mid_xlate[i], irq);
}
}
/*
* A non wait message cannot pass data or cpu source info. This current
* setup is only safe because the kernel lock owner is the only person
* who can send a message.
*
* Wrapping this whole block in a spinlock is not the safe answer either.
* A processor may get stuck with irq's off waiting to send a message and
* thus not replying to the person spinning for a reply....
*
* In the end invalidate ought to be the NMI and a very very short
* function (to avoid the old IDE disk problems), and other messages sent
* with IRQ's enabled in a civilised fashion. That will also boost
* performance.
*/
static volatile int message_cpu = NO_PROC_ID;
void smp_message_pass(int target, int msg, unsigned long data, int wait)
{
unsigned long target_map;
int p = smp_processor_id();
int irq = 15;
int i;
/* Before processors have been placed into their initial
* patterns do not send messages.
*/
if(!smp_processors_ready)
return;
/* Skip the reschedule if we are waiting to clear a
* message at this time. The reschedule cannot wait
* but is not critical.
*/
if(msg == MSG_RESCHEDULE) {
irq = 13;
if(smp_cpu_in_msg[p])
return;
}
/* Sanity check we don't re-enter this across CPU's. Only the kernel
* lock holder may send messages. For a STOP_CPU we are bringing the
* entire box to the fastest halt we can.. A reschedule carries
* no data and can occur during a flush.. guess what panic
* I got to notice this bug...
*/
if(message_cpu != NO_PROC_ID && msg != MSG_STOP_CPU && msg != MSG_RESCHEDULE) {
printk("CPU #%d: Message pass %d but pass in progress by %d of %d\n",
smp_processor_id(),msg,message_cpu, smp_msg_id);
/* I don't know how to gracefully die so that debugging
* this doesn't completely eat up my filesystems...
* let's try this...
*/
smp_cpu_in_msg[p] = 0; /* In case we come back here... */
intr_count = 0; /* and so panic don't barf... */
smp_swap(&message_cpu, NO_PROC_ID); /* push the store buffer */
sti();
printk("spinning, please L1-A, type ctrace and send output to davem\n");
while(1)
barrier();
}
smp_swap(&message_cpu, smp_processor_id()); /* store buffers... */
/* We are busy. */
smp_cpu_in_msg[p]++;
/* Reschedule is currently special. */
if(msg != MSG_RESCHEDULE) {
smp_src_cpu = p;
smp_msg_id = msg;
smp_msg_data = data;
}
#if 0
printk("SMP message pass from cpu %d to cpu %d msg %d\n", p, target, msg);
#endif
/* Set the target requirement. */
for(i = 0; i < smp_num_cpus; i++)
swap((unsigned long *) &cpu_callin_map[i], 0);
if(target == MSG_ALL_BUT_SELF) {
target_map = (cpu_present_map & ~(1<<p));
swap((unsigned long *) &cpu_callin_map[p], 1);
} else if(target == MSG_ALL) {
target_map = cpu_present_map;
} else {
for(i = 0; i < smp_num_cpus; i++)
if(i != target)
swap((unsigned long *) &cpu_callin_map[i], 1);
target_map = (1<<target);
}
/* Fire it off. */
send_ipi(target_map, irq);
switch(wait) {
case 1:
for(i = 0; i < smp_num_cpus; i++)
while(!cpu_callin_map[i])
barrier();
break;
case 2:
for(i = 0; i < smp_num_cpus; i++)
while(smp_invalidate_needed[i])
barrier();
break;
case 3:
/* For cross calls we hold message_cpu and smp_cpu_in_msg[]
* until all processors disperse. Else we have _big_ problems.
*/
return;
}
smp_cpu_in_msg[p]--;
smp_swap(&message_cpu, NO_PROC_ID);
}
struct smp_funcall {
smpfunc_t func;
unsigned long arg1;
unsigned long arg2;
unsigned long arg3;
unsigned long arg4;
unsigned long arg5;
unsigned long processors_in[NR_CPUS]; /* Set when ipi entered. */
unsigned long processors_out[NR_CPUS]; /* Set when ipi exited. */
} ccall_info;
/* Returns failure code if for example any of the cpu's failed to respond
* within a certain timeout period.
*/
#define CCALL_TIMEOUT 5000000 /* enough for initial testing */
/* #define DEBUG_CCALL */
/* Some nice day when we really thread the kernel I'd like to synchronize
* this with either a broadcast conditional variable, a resource adaptive
* generic mutex, or a convoy semaphore scheme of some sort. No reason
* we can't let multiple processors in here if the appropriate locking
* is done. Note that such a scheme assumes we will have a
* prioritized ipi scheme using different software level irq's.
*/
void smp_cross_call(smpfunc_t func, unsigned long arg1, unsigned long arg2,
unsigned long arg3, unsigned long arg4, unsigned long arg5)
{
unsigned long me = smp_processor_id();
unsigned long flags;
int i, timeout;
#ifdef DEBUG_CCALL
printk("xc%d<", me);
#endif
if(smp_processors_ready) {
save_flags(flags); cli();
if(me != active_kernel_processor)
goto cross_call_not_master;
/* Init function glue. */
ccall_info.func = func;
ccall_info.arg1 = arg1;
ccall_info.arg2 = arg2;
ccall_info.arg3 = arg3;
ccall_info.arg4 = arg4;
ccall_info.arg5 = arg5;
/* Init receive/complete mapping. */
for(i = 0; i < smp_num_cpus; i++) {
ccall_info.processors_in[i] = 0;
ccall_info.processors_out[i] = 0;
}
ccall_info.processors_in[me] = 1;
ccall_info.processors_out[me] = 1;
/* Fire it off. */
smp_message_pass(MSG_ALL_BUT_SELF, MSG_CROSS_CALL, 0, 3);
/* For debugging purposes right now we can timeout
* on both callin and callexit.
*/
timeout = CCALL_TIMEOUT;
for(i = 0; i < smp_num_cpus; i++) {
while(!ccall_info.processors_in[i] && timeout-- > 0)
barrier();
if(!ccall_info.processors_in[i])
goto procs_time_out;
}
#ifdef DEBUG_CCALL
printk("I");
#endif
/* Run local copy. */
func(arg1, arg2, arg3, arg4, arg5);
/* Spin on proc dispersion. */
timeout = CCALL_TIMEOUT;
for(i = 0; i < smp_num_cpus; i++) {
while(!ccall_info.processors_out[i] && timeout-- > 0)
barrier();
if(!ccall_info.processors_out[i])
goto procs_time_out;
}
#ifdef DEBUG_CCALL
printk("O>");
#endif
/* See wait case 3 in smp_message_pass()... */
smp_cpu_in_msg[me]--;
smp_swap(&message_cpu, NO_PROC_ID); /* store buffers... */
restore_flags(flags);
return; /* made it... */
procs_time_out:
printk("smp: Wheee, penguin drops off the bus\n");
smp_cpu_in_msg[me]--;
message_cpu = NO_PROC_ID;
restore_flags(flags);
return; /* why me... why me... */
}
/* Just need to run local copy. */
func(arg1, arg2, arg3, arg4, arg5);
return;
cross_call_not_master:
printk("Cross call initiated by non master cpu\n");
printk("akp=%x me=%08lx\n", active_kernel_processor, me);
restore_flags(flags);
panic("penguin cross call");
}
void smp_flush_cache_all(void)
{ xc0((smpfunc_t) local_flush_cache_all); }
void smp_flush_tlb_all(void)
{ xc0((smpfunc_t) local_flush_tlb_all); }
void smp_flush_cache_mm(struct mm_struct *mm)
{
if(mm->context != NO_CONTEXT)
xc1((smpfunc_t) local_flush_cache_mm, (unsigned long) mm);
}
void smp_flush_tlb_mm(struct mm_struct *mm)
{
if(mm->context != NO_CONTEXT)
xc1((smpfunc_t) local_flush_tlb_mm, (unsigned long) mm);
}
void smp_flush_cache_range(struct mm_struct *mm, unsigned long start,
unsigned long end)
{
if(mm->context != NO_CONTEXT)
xc3((smpfunc_t) local_flush_cache_range, (unsigned long) mm,
start, end);
}
void smp_flush_tlb_range(struct mm_struct *mm, unsigned long start,
unsigned long end)
{
if(mm->context != NO_CONTEXT)
xc3((smpfunc_t) local_flush_tlb_range, (unsigned long) mm,
start, end);
}
void smp_flush_cache_page(struct vm_area_struct *vma, unsigned long page)
{ xc2((smpfunc_t) local_flush_cache_page, (unsigned long) vma, page); }
void smp_flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
{ xc2((smpfunc_t) local_flush_tlb_page, (unsigned long) vma, page); }
void smp_flush_page_to_ram(unsigned long page)
{ xc1((smpfunc_t) local_flush_page_to_ram, page); }
/* Reschedule call back. */
void smp_reschedule_irq(void)
{
if(smp_processor_id() != active_kernel_processor)
panic("SMP Reschedule on CPU #%d, but #%d is active.\n",
smp_processor_id(), active_kernel_processor);
need_resched=1;
}
/* XXX FIXME: this still doesn't work right... XXX */
/* #define DEBUG_CAPTURE */
static volatile unsigned long release = 1;
static volatile int capture_level = 0;
void smp_capture(void)
{
unsigned long flags;
if(!smp_activated || !smp_commenced)
return;
#ifdef DEBUG_CAPTURE
printk("C<%d>", smp_processor_id());
#endif
save_flags(flags); cli();
if(!capture_level) {
release = 0;
smp_message_pass(MSG_ALL_BUT_SELF, MSG_CAPTURE, 0, 1);
}
capture_level++;
restore_flags(flags);
}
void smp_release(void)
{
unsigned long flags;
int i;
if(!smp_activated || !smp_commenced)
return;
#ifdef DEBUG_CAPTURE
printk("R<%d>", smp_processor_id());
#endif
save_flags(flags); cli();
if(!(capture_level - 1)) {
release = 1;
for(i = 0; i < smp_num_cpus; i++)
while(cpu_callin_map[i])
barrier();
}
capture_level -= 1;
restore_flags(flags);
}
/* Park a processor, we must watch for more IPI's to invalidate
* our cache's and TLB's. And also note we can only wait for
* "lock-less" IPI's and process those, as a result of such IPI's
* being non-maskable traps being on is enough to receive them.
*/
/* Message call back. */
void smp_message_irq(void)
{
int i=smp_processor_id();
switch(smp_msg_id) {
case MSG_CROSS_CALL:
/* Do it to it. */
ccall_info.processors_in[i] = 1;
ccall_info.func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3,
ccall_info.arg4, ccall_info.arg5);
ccall_info.processors_out[i] = 1;
break;
/*
* Halt other CPU's for a panic or reboot
*/
case MSG_STOP_CPU:
sti();
while(1)
barrier();
default:
printk("CPU #%d sent invalid cross CPU message to CPU #%d: %X(%lX).\n",
smp_src_cpu,smp_processor_id(),smp_msg_id,smp_msg_data);
break;
}
}
