758438a7b8
There's a few places where we need to fix things up for the kernel to work if it's linked at 32MB: - platforms/powermac/smp.c To start secondary cpus on pmac we patch the reset vector, which is fine. Except if we're above 32MB we don't have enough bits for an absolute branch, it needs to relative. - kernel/head_64.s - A few branches in the cpu hold code need to load the full target address and do a bctr. - after_prom_start needs to load PHYSICAL_START as the dest address, not 0. - The exception prolog needs to load the low word of the target adddress, not just the low halfword. - Fixup handling of the initial stab address. - kernel/setup_64.c smp_release_cpus() needs to write 1 to the spinloop flag near 0, not 32 MB. Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Signed-off-by: Paul Mackerras <paulus@samba.org>
877 lines
22 KiB
C
877 lines
22 KiB
C
/*
|
|
* SMP support for power macintosh.
|
|
*
|
|
* We support both the old "powersurge" SMP architecture
|
|
* and the current Core99 (G4 PowerMac) machines.
|
|
*
|
|
* Note that we don't support the very first rev. of
|
|
* Apple/DayStar 2 CPUs board, the one with the funky
|
|
* watchdog. Hopefully, none of these should be there except
|
|
* maybe internally to Apple. I should probably still add some
|
|
* code to detect this card though and disable SMP. --BenH.
|
|
*
|
|
* Support Macintosh G4 SMP by Troy Benjegerdes (hozer@drgw.net)
|
|
* and Ben Herrenschmidt <benh@kernel.crashing.org>.
|
|
*
|
|
* Support for DayStar quad CPU cards
|
|
* Copyright (C) XLR8, Inc. 1994-2000
|
|
*
|
|
* This program is free software; you can redistribute it and/or
|
|
* modify it under the terms of the GNU General Public License
|
|
* as published by the Free Software Foundation; either version
|
|
* 2 of the License, or (at your option) any later version.
|
|
*/
|
|
#include <linux/config.h>
|
|
#include <linux/kernel.h>
|
|
#include <linux/sched.h>
|
|
#include <linux/smp.h>
|
|
#include <linux/smp_lock.h>
|
|
#include <linux/interrupt.h>
|
|
#include <linux/kernel_stat.h>
|
|
#include <linux/delay.h>
|
|
#include <linux/init.h>
|
|
#include <linux/spinlock.h>
|
|
#include <linux/errno.h>
|
|
#include <linux/hardirq.h>
|
|
#include <linux/cpu.h>
|
|
#include <linux/compiler.h>
|
|
|
|
#include <asm/ptrace.h>
|
|
#include <asm/atomic.h>
|
|
#include <asm/irq.h>
|
|
#include <asm/page.h>
|
|
#include <asm/pgtable.h>
|
|
#include <asm/sections.h>
|
|
#include <asm/io.h>
|
|
#include <asm/prom.h>
|
|
#include <asm/smp.h>
|
|
#include <asm/machdep.h>
|
|
#include <asm/pmac_feature.h>
|
|
#include <asm/time.h>
|
|
#include <asm/mpic.h>
|
|
#include <asm/cacheflush.h>
|
|
#include <asm/keylargo.h>
|
|
#include <asm/pmac_low_i2c.h>
|
|
|
|
#undef DEBUG
|
|
|
|
#ifdef DEBUG
|
|
#define DBG(fmt...) udbg_printf(fmt)
|
|
#else
|
|
#define DBG(fmt...)
|
|
#endif
|
|
|
|
extern void __secondary_start_pmac_0(void);
|
|
|
|
#ifdef CONFIG_PPC32
|
|
|
|
/* Sync flag for HW tb sync */
|
|
static volatile int sec_tb_reset = 0;
|
|
|
|
/*
|
|
* Powersurge (old powermac SMP) support.
|
|
*/
|
|
|
|
/* Addresses for powersurge registers */
|
|
#define HAMMERHEAD_BASE 0xf8000000
|
|
#define HHEAD_CONFIG 0x90
|
|
#define HHEAD_SEC_INTR 0xc0
|
|
|
|
/* register for interrupting the primary processor on the powersurge */
|
|
/* N.B. this is actually the ethernet ROM! */
|
|
#define PSURGE_PRI_INTR 0xf3019000
|
|
|
|
/* register for storing the start address for the secondary processor */
|
|
/* N.B. this is the PCI config space address register for the 1st bridge */
|
|
#define PSURGE_START 0xf2800000
|
|
|
|
/* Daystar/XLR8 4-CPU card */
|
|
#define PSURGE_QUAD_REG_ADDR 0xf8800000
|
|
|
|
#define PSURGE_QUAD_IRQ_SET 0
|
|
#define PSURGE_QUAD_IRQ_CLR 1
|
|
#define PSURGE_QUAD_IRQ_PRIMARY 2
|
|
#define PSURGE_QUAD_CKSTOP_CTL 3
|
|
#define PSURGE_QUAD_PRIMARY_ARB 4
|
|
#define PSURGE_QUAD_BOARD_ID 6
|
|
#define PSURGE_QUAD_WHICH_CPU 7
|
|
#define PSURGE_QUAD_CKSTOP_RDBK 8
|
|
#define PSURGE_QUAD_RESET_CTL 11
|
|
|
|
#define PSURGE_QUAD_OUT(r, v) (out_8(quad_base + ((r) << 4) + 4, (v)))
|
|
#define PSURGE_QUAD_IN(r) (in_8(quad_base + ((r) << 4) + 4) & 0x0f)
|
|
#define PSURGE_QUAD_BIS(r, v) (PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) | (v)))
|
|
#define PSURGE_QUAD_BIC(r, v) (PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) & ~(v)))
|
|
|
|
/* virtual addresses for the above */
|
|
static volatile u8 __iomem *hhead_base;
|
|
static volatile u8 __iomem *quad_base;
|
|
static volatile u32 __iomem *psurge_pri_intr;
|
|
static volatile u8 __iomem *psurge_sec_intr;
|
|
static volatile u32 __iomem *psurge_start;
|
|
|
|
/* values for psurge_type */
|
|
#define PSURGE_NONE -1
|
|
#define PSURGE_DUAL 0
|
|
#define PSURGE_QUAD_OKEE 1
|
|
#define PSURGE_QUAD_COTTON 2
|
|
#define PSURGE_QUAD_ICEGRASS 3
|
|
|
|
/* what sort of powersurge board we have */
|
|
static int psurge_type = PSURGE_NONE;
|
|
|
|
/*
|
|
* Set and clear IPIs for powersurge.
|
|
*/
|
|
static inline void psurge_set_ipi(int cpu)
|
|
{
|
|
if (psurge_type == PSURGE_NONE)
|
|
return;
|
|
if (cpu == 0)
|
|
in_be32(psurge_pri_intr);
|
|
else if (psurge_type == PSURGE_DUAL)
|
|
out_8(psurge_sec_intr, 0);
|
|
else
|
|
PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_SET, 1 << cpu);
|
|
}
|
|
|
|
static inline void psurge_clr_ipi(int cpu)
|
|
{
|
|
if (cpu > 0) {
|
|
switch(psurge_type) {
|
|
case PSURGE_DUAL:
|
|
out_8(psurge_sec_intr, ~0);
|
|
case PSURGE_NONE:
|
|
break;
|
|
default:
|
|
PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, 1 << cpu);
|
|
}
|
|
}
|
|
}
|
|
|
|
/*
|
|
* On powersurge (old SMP powermac architecture) we don't have
|
|
* separate IPIs for separate messages like openpic does. Instead
|
|
* we have a bitmap for each processor, where a 1 bit means that
|
|
* the corresponding message is pending for that processor.
|
|
* Ideally each cpu's entry would be in a different cache line.
|
|
* -- paulus.
|
|
*/
|
|
static unsigned long psurge_smp_message[NR_CPUS];
|
|
|
|
void psurge_smp_message_recv(struct pt_regs *regs)
|
|
{
|
|
int cpu = smp_processor_id();
|
|
int msg;
|
|
|
|
/* clear interrupt */
|
|
psurge_clr_ipi(cpu);
|
|
|
|
if (num_online_cpus() < 2)
|
|
return;
|
|
|
|
/* make sure there is a message there */
|
|
for (msg = 0; msg < 4; msg++)
|
|
if (test_and_clear_bit(msg, &psurge_smp_message[cpu]))
|
|
smp_message_recv(msg, regs);
|
|
}
|
|
|
|
irqreturn_t psurge_primary_intr(int irq, void *d, struct pt_regs *regs)
|
|
{
|
|
psurge_smp_message_recv(regs);
|
|
return IRQ_HANDLED;
|
|
}
|
|
|
|
static void smp_psurge_message_pass(int target, int msg)
|
|
{
|
|
int i;
|
|
|
|
if (num_online_cpus() < 2)
|
|
return;
|
|
|
|
for (i = 0; i < NR_CPUS; i++) {
|
|
if (!cpu_online(i))
|
|
continue;
|
|
if (target == MSG_ALL
|
|
|| (target == MSG_ALL_BUT_SELF && i != smp_processor_id())
|
|
|| target == i) {
|
|
set_bit(msg, &psurge_smp_message[i]);
|
|
psurge_set_ipi(i);
|
|
}
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Determine a quad card presence. We read the board ID register, we
|
|
* force the data bus to change to something else, and we read it again.
|
|
* It it's stable, then the register probably exist (ugh !)
|
|
*/
|
|
static int __init psurge_quad_probe(void)
|
|
{
|
|
int type;
|
|
unsigned int i;
|
|
|
|
type = PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID);
|
|
if (type < PSURGE_QUAD_OKEE || type > PSURGE_QUAD_ICEGRASS
|
|
|| type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
|
|
return PSURGE_DUAL;
|
|
|
|
/* looks OK, try a slightly more rigorous test */
|
|
/* bogus is not necessarily cacheline-aligned,
|
|
though I don't suppose that really matters. -- paulus */
|
|
for (i = 0; i < 100; i++) {
|
|
volatile u32 bogus[8];
|
|
bogus[(0+i)%8] = 0x00000000;
|
|
bogus[(1+i)%8] = 0x55555555;
|
|
bogus[(2+i)%8] = 0xFFFFFFFF;
|
|
bogus[(3+i)%8] = 0xAAAAAAAA;
|
|
bogus[(4+i)%8] = 0x33333333;
|
|
bogus[(5+i)%8] = 0xCCCCCCCC;
|
|
bogus[(6+i)%8] = 0xCCCCCCCC;
|
|
bogus[(7+i)%8] = 0x33333333;
|
|
wmb();
|
|
asm volatile("dcbf 0,%0" : : "r" (bogus) : "memory");
|
|
mb();
|
|
if (type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
|
|
return PSURGE_DUAL;
|
|
}
|
|
return type;
|
|
}
|
|
|
|
static void __init psurge_quad_init(void)
|
|
{
|
|
int procbits;
|
|
|
|
if (ppc_md.progress) ppc_md.progress("psurge_quad_init", 0x351);
|
|
procbits = ~PSURGE_QUAD_IN(PSURGE_QUAD_WHICH_CPU);
|
|
if (psurge_type == PSURGE_QUAD_ICEGRASS)
|
|
PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
|
|
else
|
|
PSURGE_QUAD_BIC(PSURGE_QUAD_CKSTOP_CTL, procbits);
|
|
mdelay(33);
|
|
out_8(psurge_sec_intr, ~0);
|
|
PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, procbits);
|
|
PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
|
|
if (psurge_type != PSURGE_QUAD_ICEGRASS)
|
|
PSURGE_QUAD_BIS(PSURGE_QUAD_CKSTOP_CTL, procbits);
|
|
PSURGE_QUAD_BIC(PSURGE_QUAD_PRIMARY_ARB, procbits);
|
|
mdelay(33);
|
|
PSURGE_QUAD_BIC(PSURGE_QUAD_RESET_CTL, procbits);
|
|
mdelay(33);
|
|
PSURGE_QUAD_BIS(PSURGE_QUAD_PRIMARY_ARB, procbits);
|
|
mdelay(33);
|
|
}
|
|
|
|
static int __init smp_psurge_probe(void)
|
|
{
|
|
int i, ncpus;
|
|
|
|
/* We don't do SMP on the PPC601 -- paulus */
|
|
if (PVR_VER(mfspr(SPRN_PVR)) == 1)
|
|
return 1;
|
|
|
|
/*
|
|
* The powersurge cpu board can be used in the generation
|
|
* of powermacs that have a socket for an upgradeable cpu card,
|
|
* including the 7500, 8500, 9500, 9600.
|
|
* The device tree doesn't tell you if you have 2 cpus because
|
|
* OF doesn't know anything about the 2nd processor.
|
|
* Instead we look for magic bits in magic registers,
|
|
* in the hammerhead memory controller in the case of the
|
|
* dual-cpu powersurge board. -- paulus.
|
|
*/
|
|
if (find_devices("hammerhead") == NULL)
|
|
return 1;
|
|
|
|
hhead_base = ioremap(HAMMERHEAD_BASE, 0x800);
|
|
quad_base = ioremap(PSURGE_QUAD_REG_ADDR, 1024);
|
|
psurge_sec_intr = hhead_base + HHEAD_SEC_INTR;
|
|
|
|
psurge_type = psurge_quad_probe();
|
|
if (psurge_type != PSURGE_DUAL) {
|
|
psurge_quad_init();
|
|
/* All released cards using this HW design have 4 CPUs */
|
|
ncpus = 4;
|
|
} else {
|
|
iounmap(quad_base);
|
|
if ((in_8(hhead_base + HHEAD_CONFIG) & 0x02) == 0) {
|
|
/* not a dual-cpu card */
|
|
iounmap(hhead_base);
|
|
psurge_type = PSURGE_NONE;
|
|
return 1;
|
|
}
|
|
ncpus = 2;
|
|
}
|
|
|
|
psurge_start = ioremap(PSURGE_START, 4);
|
|
psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4);
|
|
|
|
/*
|
|
* This is necessary because OF doesn't know about the
|
|
* secondary cpu(s), and thus there aren't nodes in the
|
|
* device tree for them, and smp_setup_cpu_maps hasn't
|
|
* set their bits in cpu_possible_map and cpu_present_map.
|
|
*/
|
|
if (ncpus > NR_CPUS)
|
|
ncpus = NR_CPUS;
|
|
for (i = 1; i < ncpus ; ++i) {
|
|
cpu_set(i, cpu_present_map);
|
|
cpu_set(i, cpu_possible_map);
|
|
set_hard_smp_processor_id(i, i);
|
|
}
|
|
|
|
if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352);
|
|
|
|
return ncpus;
|
|
}
|
|
|
|
static void __init smp_psurge_kick_cpu(int nr)
|
|
{
|
|
unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8;
|
|
unsigned long a;
|
|
|
|
/* may need to flush here if secondary bats aren't setup */
|
|
for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32)
|
|
asm volatile("dcbf 0,%0" : : "r" (a) : "memory");
|
|
asm volatile("sync");
|
|
|
|
if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu", 0x353);
|
|
|
|
out_be32(psurge_start, start);
|
|
mb();
|
|
|
|
psurge_set_ipi(nr);
|
|
udelay(10);
|
|
psurge_clr_ipi(nr);
|
|
|
|
if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354);
|
|
}
|
|
|
|
/*
|
|
* With the dual-cpu powersurge board, the decrementers and timebases
|
|
* of both cpus are frozen after the secondary cpu is started up,
|
|
* until we give the secondary cpu another interrupt. This routine
|
|
* uses this to get the timebases synchronized.
|
|
* -- paulus.
|
|
*/
|
|
static void __init psurge_dual_sync_tb(int cpu_nr)
|
|
{
|
|
int t;
|
|
|
|
set_dec(tb_ticks_per_jiffy);
|
|
/* XXX fixme */
|
|
set_tb(0, 0);
|
|
last_jiffy_stamp(cpu_nr) = 0;
|
|
|
|
if (cpu_nr > 0) {
|
|
mb();
|
|
sec_tb_reset = 1;
|
|
return;
|
|
}
|
|
|
|
/* wait for the secondary to have reset its TB before proceeding */
|
|
for (t = 10000000; t > 0 && !sec_tb_reset; --t)
|
|
;
|
|
|
|
/* now interrupt the secondary, starting both TBs */
|
|
psurge_set_ipi(1);
|
|
}
|
|
|
|
static struct irqaction psurge_irqaction = {
|
|
.handler = psurge_primary_intr,
|
|
.flags = SA_INTERRUPT,
|
|
.mask = CPU_MASK_NONE,
|
|
.name = "primary IPI",
|
|
};
|
|
|
|
static void __init smp_psurge_setup_cpu(int cpu_nr)
|
|
{
|
|
|
|
if (cpu_nr == 0) {
|
|
/* If we failed to start the second CPU, we should still
|
|
* send it an IPI to start the timebase & DEC or we might
|
|
* have them stuck.
|
|
*/
|
|
if (num_online_cpus() < 2) {
|
|
if (psurge_type == PSURGE_DUAL)
|
|
psurge_set_ipi(1);
|
|
return;
|
|
}
|
|
/* reset the entry point so if we get another intr we won't
|
|
* try to startup again */
|
|
out_be32(psurge_start, 0x100);
|
|
if (setup_irq(30, &psurge_irqaction))
|
|
printk(KERN_ERR "Couldn't get primary IPI interrupt");
|
|
}
|
|
|
|
if (psurge_type == PSURGE_DUAL)
|
|
psurge_dual_sync_tb(cpu_nr);
|
|
}
|
|
|
|
void __init smp_psurge_take_timebase(void)
|
|
{
|
|
/* Dummy implementation */
|
|
}
|
|
|
|
void __init smp_psurge_give_timebase(void)
|
|
{
|
|
/* Dummy implementation */
|
|
}
|
|
|
|
/* PowerSurge-style Macs */
|
|
struct smp_ops_t psurge_smp_ops = {
|
|
.message_pass = smp_psurge_message_pass,
|
|
.probe = smp_psurge_probe,
|
|
.kick_cpu = smp_psurge_kick_cpu,
|
|
.setup_cpu = smp_psurge_setup_cpu,
|
|
.give_timebase = smp_psurge_give_timebase,
|
|
.take_timebase = smp_psurge_take_timebase,
|
|
};
|
|
#endif /* CONFIG_PPC32 - actually powersurge support */
|
|
|
|
#ifdef CONFIG_PPC64
|
|
/*
|
|
* G5s enable/disable the timebase via an i2c-connected clock chip.
|
|
*/
|
|
static struct device_node *pmac_tb_clock_chip_host;
|
|
static u8 pmac_tb_pulsar_addr;
|
|
static void (*pmac_tb_freeze)(int freeze);
|
|
static DEFINE_SPINLOCK(timebase_lock);
|
|
static unsigned long timebase;
|
|
|
|
static void smp_core99_cypress_tb_freeze(int freeze)
|
|
{
|
|
u8 data;
|
|
int rc;
|
|
|
|
/* Strangely, the device-tree says address is 0xd2, but darwin
|
|
* accesses 0xd0 ...
|
|
*/
|
|
pmac_low_i2c_setmode(pmac_tb_clock_chip_host, pmac_low_i2c_mode_combined);
|
|
rc = pmac_low_i2c_xfer(pmac_tb_clock_chip_host,
|
|
0xd0 | pmac_low_i2c_read,
|
|
0x81, &data, 1);
|
|
if (rc != 0)
|
|
goto bail;
|
|
|
|
data = (data & 0xf3) | (freeze ? 0x00 : 0x0c);
|
|
|
|
pmac_low_i2c_setmode(pmac_tb_clock_chip_host, pmac_low_i2c_mode_stdsub);
|
|
rc = pmac_low_i2c_xfer(pmac_tb_clock_chip_host,
|
|
0xd0 | pmac_low_i2c_write,
|
|
0x81, &data, 1);
|
|
|
|
bail:
|
|
if (rc != 0) {
|
|
printk("Cypress Timebase %s rc: %d\n",
|
|
freeze ? "freeze" : "unfreeze", rc);
|
|
panic("Timebase freeze failed !\n");
|
|
}
|
|
}
|
|
|
|
|
|
static void smp_core99_pulsar_tb_freeze(int freeze)
|
|
{
|
|
u8 data;
|
|
int rc;
|
|
|
|
pmac_low_i2c_setmode(pmac_tb_clock_chip_host, pmac_low_i2c_mode_combined);
|
|
rc = pmac_low_i2c_xfer(pmac_tb_clock_chip_host,
|
|
pmac_tb_pulsar_addr | pmac_low_i2c_read,
|
|
0x2e, &data, 1);
|
|
if (rc != 0)
|
|
goto bail;
|
|
|
|
data = (data & 0x88) | (freeze ? 0x11 : 0x22);
|
|
|
|
pmac_low_i2c_setmode(pmac_tb_clock_chip_host, pmac_low_i2c_mode_stdsub);
|
|
rc = pmac_low_i2c_xfer(pmac_tb_clock_chip_host,
|
|
pmac_tb_pulsar_addr | pmac_low_i2c_write,
|
|
0x2e, &data, 1);
|
|
bail:
|
|
if (rc != 0) {
|
|
printk(KERN_ERR "Pulsar Timebase %s rc: %d\n",
|
|
freeze ? "freeze" : "unfreeze", rc);
|
|
panic("Timebase freeze failed !\n");
|
|
}
|
|
}
|
|
|
|
|
|
static void smp_core99_give_timebase(void)
|
|
{
|
|
/* Open i2c bus for synchronous access */
|
|
if (pmac_low_i2c_open(pmac_tb_clock_chip_host, 0))
|
|
panic("Can't open i2c for TB sync !\n");
|
|
|
|
spin_lock(&timebase_lock);
|
|
(*pmac_tb_freeze)(1);
|
|
mb();
|
|
timebase = get_tb();
|
|
spin_unlock(&timebase_lock);
|
|
|
|
while (timebase)
|
|
barrier();
|
|
|
|
spin_lock(&timebase_lock);
|
|
(*pmac_tb_freeze)(0);
|
|
spin_unlock(&timebase_lock);
|
|
|
|
/* Close i2c bus */
|
|
pmac_low_i2c_close(pmac_tb_clock_chip_host);
|
|
}
|
|
|
|
|
|
static void __devinit smp_core99_take_timebase(void)
|
|
{
|
|
while (!timebase)
|
|
barrier();
|
|
spin_lock(&timebase_lock);
|
|
set_tb(timebase >> 32, timebase & 0xffffffff);
|
|
timebase = 0;
|
|
spin_unlock(&timebase_lock);
|
|
}
|
|
|
|
static void __init smp_core99_setup(int ncpus)
|
|
{
|
|
struct device_node *cc = NULL;
|
|
struct device_node *p;
|
|
u32 *reg;
|
|
int ok;
|
|
|
|
/* HW sync only on these platforms */
|
|
if (!machine_is_compatible("PowerMac7,2") &&
|
|
!machine_is_compatible("PowerMac7,3") &&
|
|
!machine_is_compatible("RackMac3,1"))
|
|
return;
|
|
|
|
/* Look for the clock chip */
|
|
while ((cc = of_find_node_by_name(cc, "i2c-hwclock")) != NULL) {
|
|
p = of_get_parent(cc);
|
|
ok = p && device_is_compatible(p, "uni-n-i2c");
|
|
of_node_put(p);
|
|
if (!ok)
|
|
continue;
|
|
|
|
reg = (u32 *)get_property(cc, "reg", NULL);
|
|
if (reg == NULL)
|
|
continue;
|
|
|
|
switch (*reg) {
|
|
case 0xd2:
|
|
if (device_is_compatible(cc, "pulsar-legacy-slewing")) {
|
|
pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
|
|
pmac_tb_pulsar_addr = 0xd2;
|
|
printk(KERN_INFO "Timebase clock is Pulsar chip\n");
|
|
} else if (device_is_compatible(cc, "cy28508")) {
|
|
pmac_tb_freeze = smp_core99_cypress_tb_freeze;
|
|
printk(KERN_INFO "Timebase clock is Cypress chip\n");
|
|
}
|
|
break;
|
|
case 0xd4:
|
|
pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
|
|
pmac_tb_pulsar_addr = 0xd4;
|
|
printk(KERN_INFO "Timebase clock is Pulsar chip\n");
|
|
break;
|
|
}
|
|
if (pmac_tb_freeze != NULL) {
|
|
pmac_tb_clock_chip_host = of_get_parent(cc);
|
|
of_node_put(cc);
|
|
break;
|
|
}
|
|
}
|
|
if (pmac_tb_freeze == NULL) {
|
|
smp_ops->give_timebase = smp_generic_give_timebase;
|
|
smp_ops->take_timebase = smp_generic_take_timebase;
|
|
}
|
|
}
|
|
|
|
/* nothing to do here, caches are already set up by service processor */
|
|
static inline void __devinit core99_init_caches(int cpu)
|
|
{
|
|
}
|
|
|
|
#else /* CONFIG_PPC64 */
|
|
|
|
/*
|
|
* SMP G4 powermacs use a GPIO to enable/disable the timebase.
|
|
*/
|
|
|
|
static unsigned int core99_tb_gpio; /* Timebase freeze GPIO */
|
|
|
|
static unsigned int pri_tb_hi, pri_tb_lo;
|
|
static unsigned int pri_tb_stamp;
|
|
|
|
/* not __init, called in sleep/wakeup code */
|
|
void smp_core99_give_timebase(void)
|
|
{
|
|
unsigned long flags;
|
|
unsigned int t;
|
|
|
|
/* wait for the secondary to be in take_timebase */
|
|
for (t = 100000; t > 0 && !sec_tb_reset; --t)
|
|
udelay(10);
|
|
if (!sec_tb_reset) {
|
|
printk(KERN_WARNING "Timeout waiting sync on second CPU\n");
|
|
return;
|
|
}
|
|
|
|
/* freeze the timebase and read it */
|
|
/* disable interrupts so the timebase is disabled for the
|
|
shortest possible time */
|
|
local_irq_save(flags);
|
|
pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4);
|
|
pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
|
|
mb();
|
|
pri_tb_hi = get_tbu();
|
|
pri_tb_lo = get_tbl();
|
|
pri_tb_stamp = last_jiffy_stamp(smp_processor_id());
|
|
mb();
|
|
|
|
/* tell the secondary we're ready */
|
|
sec_tb_reset = 2;
|
|
mb();
|
|
|
|
/* wait for the secondary to have taken it */
|
|
/* note: can't use udelay here, since it needs the timebase running */
|
|
for (t = 10000000; t > 0 && sec_tb_reset; --t)
|
|
barrier();
|
|
if (sec_tb_reset)
|
|
/* XXX BUG_ON here? */
|
|
printk(KERN_WARNING "Timeout waiting sync(2) on second CPU\n");
|
|
|
|
/* Now, restart the timebase by leaving the GPIO to an open collector */
|
|
pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0);
|
|
pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
|
|
local_irq_restore(flags);
|
|
}
|
|
|
|
/* not __init, called in sleep/wakeup code */
|
|
void smp_core99_take_timebase(void)
|
|
{
|
|
unsigned long flags;
|
|
|
|
/* tell the primary we're here */
|
|
sec_tb_reset = 1;
|
|
mb();
|
|
|
|
/* wait for the primary to set pri_tb_hi/lo */
|
|
while (sec_tb_reset < 2)
|
|
mb();
|
|
|
|
/* set our stuff the same as the primary */
|
|
local_irq_save(flags);
|
|
set_dec(1);
|
|
set_tb(pri_tb_hi, pri_tb_lo);
|
|
last_jiffy_stamp(smp_processor_id()) = pri_tb_stamp;
|
|
mb();
|
|
|
|
/* tell the primary we're done */
|
|
sec_tb_reset = 0;
|
|
mb();
|
|
local_irq_restore(flags);
|
|
}
|
|
|
|
/* L2 and L3 cache settings to pass from CPU0 to CPU1 on G4 cpus */
|
|
volatile static long int core99_l2_cache;
|
|
volatile static long int core99_l3_cache;
|
|
|
|
static void __devinit core99_init_caches(int cpu)
|
|
{
|
|
if (!cpu_has_feature(CPU_FTR_L2CR))
|
|
return;
|
|
|
|
if (cpu == 0) {
|
|
core99_l2_cache = _get_L2CR();
|
|
printk("CPU0: L2CR is %lx\n", core99_l2_cache);
|
|
} else {
|
|
printk("CPU%d: L2CR was %lx\n", cpu, _get_L2CR());
|
|
_set_L2CR(0);
|
|
_set_L2CR(core99_l2_cache);
|
|
printk("CPU%d: L2CR set to %lx\n", cpu, core99_l2_cache);
|
|
}
|
|
|
|
if (!cpu_has_feature(CPU_FTR_L3CR))
|
|
return;
|
|
|
|
if (cpu == 0){
|
|
core99_l3_cache = _get_L3CR();
|
|
printk("CPU0: L3CR is %lx\n", core99_l3_cache);
|
|
} else {
|
|
printk("CPU%d: L3CR was %lx\n", cpu, _get_L3CR());
|
|
_set_L3CR(0);
|
|
_set_L3CR(core99_l3_cache);
|
|
printk("CPU%d: L3CR set to %lx\n", cpu, core99_l3_cache);
|
|
}
|
|
}
|
|
|
|
static void __init smp_core99_setup(int ncpus)
|
|
{
|
|
struct device_node *cpu;
|
|
u32 *tbprop = NULL;
|
|
int i;
|
|
|
|
core99_tb_gpio = KL_GPIO_TB_ENABLE; /* default value */
|
|
cpu = of_find_node_by_type(NULL, "cpu");
|
|
if (cpu != NULL) {
|
|
tbprop = (u32 *)get_property(cpu, "timebase-enable", NULL);
|
|
if (tbprop)
|
|
core99_tb_gpio = *tbprop;
|
|
of_node_put(cpu);
|
|
}
|
|
|
|
/* XXX should get this from reg properties */
|
|
for (i = 1; i < ncpus; ++i)
|
|
smp_hw_index[i] = i;
|
|
powersave_nap = 0;
|
|
}
|
|
#endif
|
|
|
|
static int __init smp_core99_probe(void)
|
|
{
|
|
struct device_node *cpus;
|
|
int ncpus = 0;
|
|
|
|
if (ppc_md.progress) ppc_md.progress("smp_core99_probe", 0x345);
|
|
|
|
/* Count CPUs in the device-tree */
|
|
for (cpus = NULL; (cpus = of_find_node_by_type(cpus, "cpu")) != NULL;)
|
|
++ncpus;
|
|
|
|
printk(KERN_INFO "PowerMac SMP probe found %d cpus\n", ncpus);
|
|
|
|
/* Nothing more to do if less than 2 of them */
|
|
if (ncpus <= 1)
|
|
return 1;
|
|
|
|
smp_core99_setup(ncpus);
|
|
mpic_request_ipis();
|
|
core99_init_caches(0);
|
|
|
|
return ncpus;
|
|
}
|
|
|
|
static void __devinit smp_core99_kick_cpu(int nr)
|
|
{
|
|
unsigned int save_vector;
|
|
unsigned long target, flags;
|
|
volatile unsigned int *vector
|
|
= ((volatile unsigned int *)(KERNELBASE+0x100));
|
|
|
|
if (nr < 0 || nr > 3)
|
|
return;
|
|
|
|
if (ppc_md.progress)
|
|
ppc_md.progress("smp_core99_kick_cpu", 0x346);
|
|
|
|
local_irq_save(flags);
|
|
local_irq_disable();
|
|
|
|
/* Save reset vector */
|
|
save_vector = *vector;
|
|
|
|
/* Setup fake reset vector that does
|
|
* b __secondary_start_pmac_0 + nr*8 - KERNELBASE
|
|
*/
|
|
target = (unsigned long) __secondary_start_pmac_0 + nr * 8;
|
|
create_branch((unsigned long)vector, target, BRANCH_SET_LINK);
|
|
|
|
/* Put some life in our friend */
|
|
pmac_call_feature(PMAC_FTR_RESET_CPU, NULL, nr, 0);
|
|
|
|
/* FIXME: We wait a bit for the CPU to take the exception, I should
|
|
* instead wait for the entry code to set something for me. Well,
|
|
* ideally, all that crap will be done in prom.c and the CPU left
|
|
* in a RAM-based wait loop like CHRP.
|
|
*/
|
|
mdelay(1);
|
|
|
|
/* Restore our exception vector */
|
|
*vector = save_vector;
|
|
flush_icache_range((unsigned long) vector, (unsigned long) vector + 4);
|
|
|
|
local_irq_restore(flags);
|
|
if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu done", 0x347);
|
|
}
|
|
|
|
static void __devinit smp_core99_setup_cpu(int cpu_nr)
|
|
{
|
|
/* Setup L2/L3 */
|
|
if (cpu_nr != 0)
|
|
core99_init_caches(cpu_nr);
|
|
|
|
/* Setup openpic */
|
|
mpic_setup_this_cpu();
|
|
|
|
if (cpu_nr == 0) {
|
|
#ifdef CONFIG_POWER4
|
|
extern void g5_phy_disable_cpu1(void);
|
|
|
|
/* If we didn't start the second CPU, we must take
|
|
* it off the bus
|
|
*/
|
|
if (machine_is_compatible("MacRISC4") &&
|
|
num_online_cpus() < 2)
|
|
g5_phy_disable_cpu1();
|
|
#endif /* CONFIG_POWER4 */
|
|
if (ppc_md.progress) ppc_md.progress("core99_setup_cpu 0 done", 0x349);
|
|
}
|
|
}
|
|
|
|
|
|
#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PPC32)
|
|
|
|
int smp_core99_cpu_disable(void)
|
|
{
|
|
cpu_clear(smp_processor_id(), cpu_online_map);
|
|
|
|
/* XXX reset cpu affinity here */
|
|
mpic_cpu_set_priority(0xf);
|
|
asm volatile("mtdec %0" : : "r" (0x7fffffff));
|
|
mb();
|
|
udelay(20);
|
|
asm volatile("mtdec %0" : : "r" (0x7fffffff));
|
|
return 0;
|
|
}
|
|
|
|
extern void low_cpu_die(void) __attribute__((noreturn)); /* in sleep.S */
|
|
static int cpu_dead[NR_CPUS];
|
|
|
|
void cpu_die(void)
|
|
{
|
|
local_irq_disable();
|
|
cpu_dead[smp_processor_id()] = 1;
|
|
mb();
|
|
low_cpu_die();
|
|
}
|
|
|
|
void smp_core99_cpu_die(unsigned int cpu)
|
|
{
|
|
int timeout;
|
|
|
|
timeout = 1000;
|
|
while (!cpu_dead[cpu]) {
|
|
if (--timeout == 0) {
|
|
printk("CPU %u refused to die!\n", cpu);
|
|
break;
|
|
}
|
|
msleep(1);
|
|
}
|
|
cpu_dead[cpu] = 0;
|
|
}
|
|
|
|
#endif
|
|
|
|
/* Core99 Macs (dual G4s and G5s) */
|
|
struct smp_ops_t core99_smp_ops = {
|
|
.message_pass = smp_mpic_message_pass,
|
|
.probe = smp_core99_probe,
|
|
.kick_cpu = smp_core99_kick_cpu,
|
|
.setup_cpu = smp_core99_setup_cpu,
|
|
.give_timebase = smp_core99_give_timebase,
|
|
.take_timebase = smp_core99_take_timebase,
|
|
#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PPC32)
|
|
.cpu_disable = smp_core99_cpu_disable,
|
|
.cpu_die = smp_core99_cpu_die,
|
|
#endif
|
|
};
|